tcp_bbr: add bbr_check_probe_rtt_done() helper

This patch add a helper function bbr_check_probe_rtt_done() to
  1. check the condition to see if bbr should exit probe_rtt mode;
  2. process the logic of exiting probe_rtt mode.

Fixes: 0f8782ea1497 ("tcp_bbr: add BBR congestion control")
Signed-off-by: Kevin Yang <yyd@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Reviewed-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ipv4: tcp: send zero IPID for RST and ACK sent in SYN-RECV and TIME-WAIT state

tcp uses per-cpu (and per namespace) sockets (net->ipv4.tcp_sk) internally
to send some control packets.

1) RST packets, through tcp_v4_send_reset()
2) ACK packets in SYN-RECV and TIME-WAIT state, through tcp_v4_send_ack()

These packets assert IP_DF, and also use the hashed IP ident generator
to provide an IPv4 ID number.

Geoff Alexander reported this could be used to build off-path attacks.

These packets should not be fragmented, since their size is smaller than
IPV4_MIN_MTU. Only some tunneled paths could eventually have to fragment,
regardless of inner IPID.

We really can use zero IPID, to address the flaw, and as a bonus,
avoid a couple of atomic operations in ip_idents_reserve()

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Geoff Alexander <alexandg@cs.unm.edu>
Tested-by: Geoff Alexander <alexandg@cs.unm.edu>
Signed-off-by: David S. Miller <davem@davemloft.net>

addrconf: reduce unnecessary atomic allocations

All the 3 callers of addrconf_add_mroute() assert RTNL
lock, they don't take any additional lock either, so
it is safe to convert it to GFP_KERNEL.

Same for sit_add_v4_addrs().

Cc: David Ahern <dsahern@gmail.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net_sched: fix unused variable warning in stmmac

The new tcf_exts_for_each_action() macro doesn't reference its
arguments when CONFIG_NET_CLS_ACT is disabled, which leads to
a harmless warning in at least one driver:

drivers/net/ethernet/stmicro/stmmac/stmmac_tc.c: In function 'tc_fill_actions':
drivers/net/ethernet/stmicro/stmmac/stmmac_tc.c:64:6: error: unused variable 'i' [-Werror=unused-variable]

Adding a cast to void lets us avoid this kind of warning.
To be on the safe side, do it for all three arguments, not
just the one that caused the warning.

Fixes: 244cd96adb5f ("net_sched: remove list_head from tc_action")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

sch_cake: Fix TC filter flow override and expand it to hosts as well

The TC filter flow mapping override completely skipped the call to
cake_hash(); however that meant that the internal state was not being
updated, which ultimately leads to deadlocks in some configurations. Fix
that by passing the overridden flow ID into cake_hash() instead so it can
react appropriately.

In addition, the major number of the class ID can now be set to override
the host mapping in host isolation mode. If both host and flow are
overridden (or if the respective modes are disabled), flow dissection and
hashing will be skipped entirely; otherwise, the hashing will be kept for
the portions that are not set by the filter.

Signed-off-by: Toke Høiland-Jørgensen <toke@toke.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ncsi: Fixup .dumpit message flags and ID check in Netlink handler

The ncsi_pkg_info_all_nl() .dumpit handler is missing the NLM_F_MULTI
flag, causing additional package information after the first to be lost.
Also fixup a sanity check in ncsi_write_package_info() to reject out of
range package IDs.

Signed-off-by: Samuel Mendoza-Jonas <sam@mendozajonas.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: ethernet: renesas: use SPDX identifier for Renesas drivers

Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
Acked-by: Sergei Shtylyov <sergei.shtylyov@cogentembedded.com>
Reviewed-by: Simon Horman <horms+renesas@verge.net.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'net_sched-fixes'

Cong Wang says:

====================
net_sched: pending clean up and bug fixes

This patchset aims to clean up and fixes some bugs in current
merge window, this is why it is targeting -net.

Patch 1-5 are clean up Vlad's patches merged in current merge
window, patch 6 is just a trivial cleanup.

Patch 7 reverts a lockdep warning fix and patch 8 provides a better
fix for it.

Patch 9 fixes a potential deadlock found by me during code review.

Please see each patch for details.
====================

Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>

act_ife: fix a potential deadlock

use_all_metadata() acquires read_lock(&ife_mod_lock), then calls
add_metainfo() which calls find_ife_oplist() which acquires the same
lock again. Deadlock!

Introduce __add_metainfo() which accepts struct tcf_meta_ops *ops
as an additional parameter and let its callers to decide how
to find it. For use_all_metadata(), it already has ops, no
need to find it again, just call __add_metainfo() directly.

And, as ife_mod_lock is only needed for find_ife_oplist(),
this means we can make non-atomic allocation for populate_metalist()
now.

Fixes: 817e9f2c5c26 ("act_ife: acquire ife_mod_lock before reading ifeoplist")
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

act_ife: move tcfa_lock down to where necessary

The only time we need to take tcfa_lock is when adding
a new metainfo to an existing ife->metalist. We don't need
to take tcfa_lock so early and so broadly in tcf_ife_init().

This means we can always take ife_mod_lock first, avoid the
reverse locking ordering warning as reported by Vlad.

Reported-by: Vlad Buslov <vladbu@mellanox.com>
Tested-by: Vlad Buslov <vladbu@mellanox.com>
Cc: Vlad Buslov <vladbu@mellanox.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Revert "net: sched: act_ife: disable bh when taking ife_mod_lock"

This reverts commit 42c625a486f3 ("net: sched: act_ife: disable bh
when taking ife_mod_lock"), because what ife_mod_lock protects
is absolutely not touched in rate est timer BH context, they have
no race.

A better fix is following up.

Cc: Vlad Buslov <vladbu@mellanox.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net_sched: remove unused tcfa_capab

Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net_sched: remove list_head from tc_action

After commit 90b73b77d08e, list_head is no longer needed.
Now we just need to convert the list iteration to array
iteration for drivers.

Fixes: 90b73b77d08e ("net: sched: change action API to use array of pointers to actions")
Cc: Jiri Pirko <jiri@mellanox.com>
Cc: Vlad Buslov <vladbu@mellanox.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net_sched: remove unused tcf_idr_check()

tcf_idr_check() is replaced by tcf_idr_check_alloc(),
and __tcf_idr_check() now can be folded into tcf_idr_search().

Fixes: 0190c1d452a9 ("net: sched: atomically check-allocate action")
Cc: Jiri Pirko <jiri@mellanox.com>
Cc: Vlad Buslov <vladbu@mellanox.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net_sched: remove unused parameter for tcf_action_delete()

Fixes: 16af6067392c ("net: sched: implement reference counted action release")
Cc: Jiri Pirko <jiri@mellanox.com>
Cc: Vlad Buslov <vladbu@mellanox.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net_sched: remove unnecessary ops->delete()

All ops->delete() wants is getting the tn->idrinfo, but we already
have tc_action before calling ops->delete(), and tc_action has
a pointer ->idrinfo.

More importantly, each type of action does the same thing, that is,
just calling tcf_idr_delete_index().

So it can be just removed.

Fixes: b409074e6693 ("net: sched: add 'delete' function to action ops")
Cc: Jiri Pirko <jiri@mellanox.com>
Cc: Vlad Buslov <vladbu@mellanox.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net_sched: improve and refactor tcf_action_put_many()

tcf_action_put_many() is mostly called to clean up actions on
failure path, but tcf_action_put_many(&actions[acts_deleted]) is
used in the ugliest way: it passes a slice of the array and
uses an additional NULL at the end to avoid out-of-bound
access.

acts_deleted is completely unnecessary since we can teach
tcf_action_put_many() scan the whole array and checks against
NULL pointer. Which also means tcf_action_delete() should
set deleted action pointers to NULL to avoid double free.

Fixes: 90b73b77d08e ("net: sched: change action API to use array of pointers to actions")
Cc: Jiri Pirko <jiri@mellanox.com>
Cc: Vlad Buslov <vladbu@mellanox.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

hv_netvsc: ignore devices that are not PCI

Registering another device with same MAC address (such as TAP, VPN or
DPDK KNI) will confuse the VF autobinding logic.  Restrict the search
to only run if the device is known to be a PCI attached VF.

Fixes: e8ff40d4bff1 ("hv_netvsc: improve VF device matching")
Signed-off-by: Stephen Hemminger <sthemmin@microsoft.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

rds: tcp: remove duplicated include from tcp.c

Remove duplicated include.

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Acked-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

sch_cake: Remove unused including <linux/version.h>

Remove including <linux/version.h> that don't need it.

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

rhashtable: remove duplicated include from rhashtable.c

Remove duplicated include.

Signed-off-by: Yue Haibing <yuehaibing@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ipv6: Put lwtstate when destroying fib6_info

Prior to the introduction of fib6_info lwtstate was managed by the dst
code. With fib6_info releasing lwtstate needs to be done when the struct
is freed.

Fixes: 93531c674315 ("net/ipv6: separate handling of FIB entries from dst based routes")
Signed-off-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

r8152: disable RX aggregation on new Dell TB16 dock

There's a new Dell TB16 dock with a different iSerialNumber.

Apply the same fix from commit 0b1655143df0 ("r8152: disable RX
aggregation on Dell TB16 dock") to this model.

BugLink: https://bugs.launchpad.net/bugs/1785780
Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge branch 'qed-Misc-fixes-in-the-interface-with-the-MFW'

Tomer Tayar says:

====================
qed: Misc fixes in the interface with the MFW

This patch series fixes several issues in the driver's interface with the
management FW (MFW).

v1->v2:
- Fix loop counter decrement to be pre instead of post.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

qed: Avoid sending mailbox commands when MFW is not responsive

Keep sending mailbox commands to the MFW when it is not responsive ends up
with a redundant amount of timeout expiries.
This patch prints the MCP status on the first command which is not
responded, and blocks the following commands.
Since the (un)load request commands might be not responded due to other
PFs, the patch also adds the option to skip the blocking upon a failure.

Signed-off-by: Tomer Tayar <Tomer.Tayar@cavium.com>
Signed-off-by: Ariel Elior <Ariel.Elior@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qed: Prevent a possible deadlock during driver load and unload

The MFW manages an internal lock to prevent concurrent hardware
(de)initialization of different PFs.
This, together with the busy-waiting for the MFW's responses for commands,
might lead to a deadlock during concurrent load or unload of PFs.
This patch adds the option to sleep within the busy-waiting, and uses it
for the (un)load requests (which are not sent from an interrupt context) to
prevent the possible deadlock.

Signed-off-by: Tomer Tayar <Tomer.Tayar@cavium.com>
Signed-off-by: Ariel Elior <Ariel.Elior@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qed: Wait for MCP halt and resume commands to take place

Successive iterations of halting and resuming the management chip (MCP)
might fail, since currently the driver doesn't wait for these operations to
actually take place.
This patch prevents the driver from moving forward before the operations
are reflected in the state register.

Signed-off-by: Tomer Tayar <Tomer.Tayar@cavium.com>
Signed-off-by: Ariel Elior <Ariel.Elior@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

qed: Wait for ready indication before rereading the shmem

The MFW might be reset and re-update its shared memory.
Upon the detection of such a reset the driver rereads this memory, but it
has to wait till the data is valid.
This patch adds the missing wait for a data ready indication.

Signed-off-by: Tomer Tayar <Tomer.Tayar@cavium.com>
Signed-off-by: Ariel Elior <Ariel.Elior@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ip6_vti: fix a null pointer deference when destroy vti6 tunnel

If load ip6_vti module and create a network namespace when set
fb_tunnels_only_for_init_net to 1, then exit the namespace will
cause following crash:

[ 6601.677036] BUG: unable to handle kernel NULL pointer dereference at 0000000000000008
[ 6601.679057] PGD 8000000425eca067 P4D 8000000425eca067 PUD 424292067 PMD 0
[ 6601.680483] Oops: 0000 [#1] SMP PTI
[ 6601.681223] CPU: 7 PID: 93 Comm: kworker/u16:1 Kdump: loaded Tainted: G            E     4.18.0+ #3
[ 6601.683153] Hardware name: Fedora Project OpenStack Nova, BIOS seabios-1.7.5-11.el7 04/01/2014
[ 6601.684919] Workqueue: netns cleanup_net
[ 6601.685742] RIP: 0010:vti6_exit_batch_net+0x87/0xd0 [ip6_vti]
[ 6601.686932] Code: 7b 08 48 89 e6 e8 b9 ea d3 dd 48 8b 1b 48 85 db 75 ec 48 83 c5 08 48 81 fd 00 01 00 00 75 d5 49 8b 84 24 08 01 00 00 48 89 e6 <48> 8b 78 08 e8 90 ea d3 dd 49 8b 45 28 49 39 c6 4c 8d 68 d8 75 a1
[ 6601.690735] RSP: 0018:ffffa897c2737de0 EFLAGS: 00010246
[ 6601.691846] RAX: 0000000000000000 RBX: 0000000000000000 RCX: dead000000000200
[ 6601.693324] RDX: 0000000000000015 RSI: ffffa897c2737de0 RDI: ffffffff9f2ea9e0
[ 6601.694824] RBP: 0000000000000100 R08: 0000000000000000 R09: 0000000000000000
[ 6601.696314] R10: 0000000000000001 R11: 0000000000000000 R12: ffff8dc323c07e00
[ 6601.697812] R13: ffff8dc324a63100 R14: ffffa897c2737e30 R15: ffffa897c2737e30
[ 6601.699345] FS:  0000000000000000(0000) GS:ffff8dc33fdc0000(0000) knlGS:0000000000000000
[ 6601.701068] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 6601.702282] CR2: 0000000000000008 CR3: 0000000424966002 CR4: 00000000001606e0
[ 6601.703791] Call Trace:
[ 6601.704329]  cleanup_net+0x1b4/0x2c0
[ 6601.705268]  process_one_work+0x16c/0x370
[ 6601.706145]  worker_thread+0x49/0x3e0
[ 6601.706942]  kthread+0xf8/0x130
[ 6601.707626]  ? rescuer_thread+0x340/0x340
[ 6601.708476]  ? kthread_bind+0x10/0x10
[ 6601.709266]  ret_from_fork+0x35/0x40

Reproduce:
modprobe ip6_vti
echo 1 > /proc/sys/net/core/fb_tunnels_only_for_init_net
unshare -n
exit

This because ip6n->tnls_wc[0] point to fallback device in default, but
in non-default namespace, ip6n->tnls_wc[0] will be NULL, so add the NULL
check comparatively.

Fixes: e2948e5af8ee ("ip6_vti: fix creating fallback tunnel device for vti6")
Signed-off-by: Haishuang Yan <yanhaishuang@cmss.chinamobile.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) Fix races in IPVS, from Tan Hu.

 2) Missing unbind in matchall classifier, from Hangbin Liu.

 3) Missing act_ife action release, from Vlad Buslov.

 4) Cure lockdep splats in ila, from Cong Wang.

 5) veth queue leak on link delete, from Toshiaki Makita.

 6) Disable isdn's IIOCDBGVAR ioctl, it exposes kernel addresses. From
    Kees Cook.

 7) RCU usage fixup in XDP, from Tariq Toukan.

 8) Two TCP ULP fixes from Daniel Borkmann.

 9) r8169 needs REALTEK_PHY as a Kconfig dependency, from Heiner
    Kallweit.

10) Always take tcf_lock with BH disabled, otherwise we can deadlock
    with rate estimator code paths. From Vlad Buslov.

11) Don't use MSI-X on RTL8106e r8169 chips, they don't resume properly.
    From Jian-Hong Pan.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (41 commits)
  ip6_vti: fix creating fallback tunnel device for vti6
  ip_vti: fix a null pointer deferrence when create vti fallback tunnel
  r8169: don't use MSI-X on RTL8106e
  net: lan743x_ptp: convert to ktime_get_clocktai_ts64
  net: sched: always disable bh when taking tcf_lock
  ip6_vti: simplify stats handling in vti6_xmit
  bpf: fix redirect to map under tail calls
  r8169: add missing Kconfig dependency
  tools/bpf: fix bpf selftest test_cgroup_storage failure
  bpf, sockmap: fix sock_map_ctx_update_elem race with exist/noexist
  bpf, sockmap: fix map elem deletion race with smap_stop_sock
  bpf, sockmap: fix leakage of smap_psock_map_entry
  tcp, ulp: fix leftover icsk_ulp_ops preventing sock from reattach
  tcp, ulp: add alias for all ulp modules
  bpf: fix a rcu usage warning in bpf_prog_array_copy_core()
  samples/bpf: all XDP samples should unload xdp/bpf prog on SIGTERM
  net/xdp: Fix suspicious RCU usage warning
  net/mlx5e: Delete unneeded function argument
  Documentation: networking: ti-cpsw: correct cbs parameters for Eth1 100Mb
  isdn: Disable IIOCDBGVAR
  ...

ip6_vti: fix creating fallback tunnel device for vti6

When set fb_tunnels_only_for_init_net to 1, don't create fallback tunnel
device for vti6 when a new namespace is created.

Tested:
[root@builder2 ~]# modprobe ip6_tunnel
[root@builder2 ~]# modprobe ip6_vti
[root@builder2 ~]# echo 1 > /proc/sys/net/core/fb_tunnels_only_for_init_net
[root@builder2 ~]# unshare -n
[root@builder2 ~]# ip link
1: lo: <LOOPBACK> mtu 65536 qdisc noop state DOWN mode DEFAULT group
default qlen 1000
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00

Signed-off-by: Haishuang Yan <yanhaishuang@cmss.chinamobile.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

ip_vti: fix a null pointer deferrence when create vti fallback tunnel

After set fb_tunnels_only_for_init_net to 1, the itn->fb_tunnel_dev will
be NULL and will cause following crash:

[ 2742.849298] BUG: unable to handle kernel NULL pointer dereference at 0000000000000941
[ 2742.851380] PGD 800000042c21a067 P4D 800000042c21a067 PUD 42aaed067 PMD 0
[ 2742.852818] Oops: 0002 [#1] SMP PTI
[ 2742.853570] CPU: 7 PID: 2484 Comm: unshare Kdump: loaded Not tainted 4.18.0-rc8+ #2
[ 2742.855163] Hardware name: Fedora Project OpenStack Nova, BIOS seabios-1.7.5-11.el7 04/01/2014
[ 2742.856970] RIP: 0010:vti_init_net+0x3a/0x50 [ip_vti]
[ 2742.858034] Code: 90 83 c0 48 c7 c2 20 a1 83 c0 48 89 fb e8 6e 3b f6 ff 85 c0 75 22 8b 0d f4 19 00 00 48 8b 93 00 14 00 00 48 8b 14 ca 48 8b 12 <c6> 82 41 09 00 00 04 c6 82 38 09 00 00 45 5b c3 66 0f 1f 44 00 00
[ 2742.861940] RSP: 0018:ffff9be28207fde0 EFLAGS: 00010246
[ 2742.863044] RAX: 0000000000000000 RBX: ffff8a71ebed4980 RCX: 0000000000000013
[ 2742.864540] RDX: 0000000000000000 RSI: 0000000000000013 RDI: ffff8a71ebed4980
[ 2742.866020] RBP: ffff8a71ea717000 R08: ffffffffc083903c R09: ffff8a71ea717000
[ 2742.867505] R10: 0000000000000000 R11: 0000000000000000 R12: ffff8a71ebed4980
[ 2742.868987] R13: 0000000000000013 R14: ffff8a71ea5b49c0 R15: 0000000000000000
[ 2742.870473] FS:  00007f02266c9740(0000) GS:ffff8a71ffdc0000(0000) knlGS:0000000000000000
[ 2742.872143] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 2742.873340] CR2: 0000000000000941 CR3: 000000042bc20006 CR4: 00000000001606e0
[ 2742.874821] Call Trace:
[ 2742.875358]  ops_init+0x38/0xf0
[ 2742.876078]  setup_net+0xd9/0x1f0
[ 2742.876789]  copy_net_ns+0xb7/0x130
[ 2742.877538]  create_new_namespaces+0x11a/0x1d0
[ 2742.878525]  unshare_nsproxy_namespaces+0x55/0xa0
[ 2742.879526]  ksys_unshare+0x1a7/0x330
[ 2742.880313]  __x64_sys_unshare+0xe/0x20
[ 2742.881131]  do_syscall_64+0x5b/0x180
[ 2742.881933]  entry_SYSCALL_64_after_hwframe+0x44/0xa9

Reproduce:
echo 1 > /proc/sys/net/core/fb_tunnels_only_for_init_net
modprobe ip_vti
unshare -n

Fixes: 79134e6ce2c9 ("net: do not create fallback tunnels for non-default namespaces")
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Haishuang Yan <yanhaishuang@cmss.chinamobile.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

r8169: don't use MSI-X on RTL8106e

Found the ethernet network on ASUS X441UAR doesn't come back on resume
from suspend when using MSI-X.  The chip is RTL8106e - version 39.

[   21.848357] libphy: r8169: probed
[   21.848473] r8169 0000:02:00.0 eth0: RTL8106e, 0c:9d:92:32:67:b4, XID
44900000, IRQ 127
[   22.518860] r8169 0000:02:00.0 enp2s0: renamed from eth0
[   29.458041] Generic PHY r8169-200:00: attached PHY driver [Generic
PHY] (mii_bus:phy_addr=r8169-200:00, irq=IGNORE)
[   63.227398] r8169 0000:02:00.0 enp2s0: Link is Up - 100Mbps/Full -
flow control off
[  124.514648] Generic PHY r8169-200:00: attached PHY driver [Generic
PHY] (mii_bus:phy_addr=r8169-200:00, irq=IGNORE)

Here is the ethernet controller in detail:

02:00.0 Ethernet controller [0200]: Realtek Semiconductor Co., Ltd.
RTL8101/2/6E PCI Express Fast/Gigabit Ethernet controller [10ec:8136]
(rev 07)
Subsystem: ASUSTeK Computer Inc. RTL810xE PCI Express Fast
Ethernet controller [1043:200f]
Flags: bus master, fast devsel, latency 0, IRQ 16
I/O ports at e000 [size=256]
Memory at ef100000 (64-bit, non-prefetchable) [size=4K]
Memory at e0000000 (64-bit, prefetchable) [size=16K]
Capabilities: <access denied>
Kernel driver in use: r8169
Kernel modules: r8169

Falling back to MSI fixes the issue.

Fixes: 6c6aa15fdea5 ("r8169: improve interrupt handling")
Signed-off-by: Jian-Hong Pan <jian-hong@endlessm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: lan743x_ptp: convert to ktime_get_clocktai_ts64

timekeeping_clocktai64() has been renamed to ktime_get_clocktai_ts64()
for consistency with the other ktime_get_* access functions.

Rename the new caller that has come up as well.

Question: this is the only ptp driver that sets the hardware time
to the current system time in TAI. Why does it do that?

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>

net: sched: always disable bh when taking tcf_lock

Recently, ops->init() and ops->dump() of all actions were modified to
always obtain tcf_lock when accessing private action state. Actions that
don't depend on tcf_lock for synchronization with their data path use
non-bh locking API. However, tcf_lock is also used to protect rate
estimator stats in softirq context by timer callback.

Change ops->init() and ops->dump() of all actions to disable bh when using
tcf_lock to prevent deadlock reported by following lockdep warning:

[  105.470398] ================================
[  105.475014] WARNING: inconsistent lock state
[  105.479628] 4.18.0-rc8+ #664 Not tainted
[  105.483897] --------------------------------
[  105.488511] inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
[  105.494871] swapper/16/0 [HC0[0]:SC1[1]:HE1:SE0] takes:
[  105.500449] 00000000f86c012e (&(&p->tcfa_lock)->rlock){+.?.}, at: est_fetch_counters+0x3c/0xa0
[  105.509696] {SOFTIRQ-ON-W} state was registered at:
[  105.514925]   _raw_spin_lock+0x2c/0x40
[  105.519022]   tcf_bpf_init+0x579/0x820 [act_bpf]
[  105.523990]   tcf_action_init_1+0x4e4/0x660
[  105.528518]   tcf_action_init+0x1ce/0x2d0
[  105.532880]   tcf_exts_validate+0x1d8/0x200
[  105.537416]   fl_change+0x55a/0x268b [cls_flower]
[  105.542469]   tc_new_tfilter+0x748/0xa20
[  105.546738]   rtnetlink_rcv_msg+0x56a/0x6d0
[  105.551268]   netlink_rcv_skb+0x18d/0x200
[  105.555628]   netlink_unicast+0x2d0/0x370
[  105.559990]   netlink_sendmsg+0x3b9/0x6a0
[  105.564349]   sock_sendmsg+0x6b/0x80
[  105.568271]   ___sys_sendmsg+0x4a1/0x520
[  105.572547]   __sys_sendmsg+0xd7/0x150
[  105.576655]   do_syscall_64+0x72/0x2c0
[  105.580757]   entry_SYSCALL_64_after_hwframe+0x49/0xbe
[  105.586243] irq event stamp: 489296
[  105.590084] hardirqs last  enabled at (489296): [<ffffffffb507e639>] _raw_spin_unlock_irq+0x29/0x40
[  105.599765] hardirqs last disabled at (489295): [<ffffffffb507e745>] _raw_spin_lock_irq+0x15/0x50
[  105.609277] softirqs last  enabled at (489292): [<ffffffffb413a6a3>] irq_enter+0x83/0xa0
[  105.618001] softirqs last disabled at (489293): [<ffffffffb413a800>] irq_exit+0x140/0x190
[  105.626813]
               other info that might help us debug this:
[  105.633976]  Possible unsafe locking scenario:

[  105.640526]        CPU0
[  105.643325]        ----
[  105.646125]   lock(&(&p->tcfa_lock)->rlock);
[  105.650747]   <Interrupt>
[  105.653717]     lock(&(&p->tcfa_lock)->rlock);
[  105.658514]
                *** DEADLOCK ***

[  105.665349] 1 lock held by swapper/16/0:
[  105.669629]  #0: 00000000a640ad99 ((&est->timer)){+.-.}, at: call_timer_fn+0x10b/0x550
[  105.678200]
               stack backtrace:
[  105.683194] CPU: 16 PID: 0 Comm: swapper/16 Not tainted 4.18.0-rc8+ #664
[  105.690249] Hardware name: Supermicro SYS-2028TP-DECR/X10DRT-P, BIOS 2.0b 03/30/2017
[  105.698626] Call Trace:
[  105.701421]  <IRQ>
[  105.703791]  dump_stack+0x92/0xeb
[  105.707461]  print_usage_bug+0x336/0x34c
[  105.711744]  mark_lock+0x7c9/0x980
[  105.715500]  ? print_shortest_lock_dependencies+0x2e0/0x2e0
[  105.721424]  ? check_usage_forwards+0x230/0x230
[  105.726315]  __lock_acquire+0x923/0x26f0
[  105.730597]  ? debug_show_all_locks+0x240/0x240
[  105.735478]  ? mark_lock+0x493/0x980
[  105.739412]  ? check_chain_key+0x140/0x1f0
[  105.743861]  ? __lock_acquire+0x836/0x26f0
[  105.748323]  ? lock_acquire+0x12e/0x290
[  105.752516]  lock_acquire+0x12e/0x290
[  105.756539]  ? est_fetch_counters+0x3c/0xa0
[  105.761084]  _raw_spin_lock+0x2c/0x40
[  105.765099]  ? est_fetch_counters+0x3c/0xa0
[  105.769633]  est_fetch_counters+0x3c/0xa0
[  105.773995]  est_timer+0x87/0x390
[  105.777670]  ? est_fetch_counters+0xa0/0xa0
[  105.782210]  ? lock_acquire+0x12e/0x290
[  105.786410]  call_timer_fn+0x161/0x550
[  105.790512]  ? est_fetch_counters+0xa0/0xa0
[  105.795055]  ? del_timer_sync+0xd0/0xd0
[  105.799249]  ? __lock_is_held+0x93/0x110
[  105.803531]  ? mark_held_locks+0x20/0xe0
[  105.807813]  ? _raw_spin_unlock_irq+0x29/0x40
[  105.812525]  ? est_fetch_counters+0xa0/0xa0
[  105.817069]  ? est_fetch_counters+0xa0/0xa0
[  105.821610]  run_timer_softirq+0x3c4/0x9f0
[  105.826064]  ? lock_acquire+0x12e/0x290
[  105.830257]  ? __bpf_trace_timer_class+0x10/0x10
[  105.835237]  ? __lock_is_held+0x25/0x110
[  105.839517]  __do_softirq+0x11d/0x7bf
[  105.843542]  irq_exit+0x140/0x190
[  105.847208]  smp_apic_timer_interrupt+0xac/0x3b0
[  105.852182]  apic_timer_interrupt+0xf/0x20
[  105.856628]  </IRQ>
[  105.859081] RIP: 0010:cpuidle_enter_state+0xd8/0x4d0
[  105.864395] Code: 46 ff 48 89 44 24 08 0f 1f 44 00 00 31 ff e8 cf ec 46 ff 80 7c 24 07 00 0f 85 1d 02 00 00 e8 9f 90 4b ff fb 66 0f 1f 44 00 00 <4c> 8b 6c 24 08 4d 29 fd 0f 80 36 03 00 00 4c 89 e8 48 ba cf f7 53
[  105.884288] RSP: 0018:ffff8803ad94fd20 EFLAGS: 00000246 ORIG_RAX: ffffffffffffff13
[  105.892494] RAX: 0000000000000000 RBX: ffffe8fb300829c0 RCX: ffffffffb41e19e1
[  105.899988] RDX: 0000000000000007 RSI: dffffc0000000000 RDI: ffff8803ad9358ac
[  105.907503] RBP: ffffffffb6636300 R08: 0000000000000004 R09: 0000000000000000
[  105.914997] R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000004
[  105.922487] R13: ffffffffb6636140 R14: ffffffffb66362d8 R15: 000000188d36091b
[  105.929988]  ? trace_hardirqs_on_caller+0x141/0x2d0
[  105.935232]  do_idle+0x28e/0x320
[  105.938817]  ? arch_cpu_idle_exit+0x40/0x40
[  105.943361]  ? mark_lock+0x8c1/0x980
[  105.947295]  ? _raw_spin_unlock_irqrestore+0x32/0x60
[  105.952619]  cpu_startup_entry+0xc2/0xd0
[  105.956900]  ? cpu_in_idle+0x20/0x20
[  105.960830]  ? _raw_spin_unlock_irqrestore+0x32/0x60
[  105.966146]  ? trace_hardirqs_on_caller+0x141/0x2d0
[  105.971391]  start_secondary+0x2b5/0x360
[  105.975669]  ? set_cpu_sibling_map+0x1330/0x1330
[  105.980654]  secondary_startup_64+0xa5/0xb0

Taking tcf_lock in sample action with bh disabled causes lockdep to issue a
warning regarding possible irq lock inversion dependency between tcf_lock,
and psample_groups_lock that is taken when holding tcf_lock in sample init:

[  162.108959]  Possible interrupt unsafe locking scenario:

[  162.116386]        CPU0                    CPU1
[  162.121277]        ----                    ----
[  162.126162]   lock(psample_groups_lock);
[  162.130447]                                local_irq_disable();
[  162.136772]                                lock(&(&p->tcfa_lock)->rlock);
[  162.143957]                                lock(psample_groups_lock);
[  162.150813]   <Interrupt>
[  162.153808]     lock(&(&p->tcfa_lock)->rlock);
[  162.158608]
                *** DEADLOCK ***

In order to prevent potential lock inversion dependency between tcf_lock
and psample_groups_lock, extract call to psample_group_get() from tcf_lock
protected section in sample action init function.

Fixes: 4e232818bd32 ("net: sched: act_mirred: remove dependency on rtnl lock")
Fixes: 764e9a24480f ("net: sched: act_vlan: remove dependency on rtnl lock")
Fixes: 729e01260989 ("net: sched: act_tunnel_key: remove dependency on rtnl lock")
Fixes: d77284956656 ("net: sched: act_sample: remove dependency on rtnl lock")
Fixes: e8917f437006 ("net: sched: act_gact: remove dependency on rtnl lock")
Fixes: b6a2b971c0b0 ("net: sched: act_csum: remove dependency on rtnl lock")
Fixes: 2142236b4584 ("net: sched: act_bpf: remove dependency on rtnl lock")
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Merge tag 'for-linus' of git://git./virt/kvm/kvm

Pull first set of KVM updates from Paolo Bonzini:
 "PPC:
   - minor code cleanups

  x86:
   - PCID emulation and CR3 caching for shadow page tables
   - nested VMX live migration
   - nested VMCS shadowing
   - optimized IPI hypercall
   - some optimizations

  ARM will come next week"

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (85 commits)
  kvm: x86: Set highest physical address bits in non-present/reserved SPTEs
  KVM/x86: Use CC_SET()/CC_OUT in arch/x86/kvm/vmx.c
  KVM: X86: Implement PV IPIs in linux guest
  KVM: X86: Add kvm hypervisor init time platform setup callback
  KVM: X86: Implement "send IPI" hypercall
  KVM/x86: Move X86_CR4_OSXSAVE check into kvm_valid_sregs()
  KVM: x86: Skip pae_root shadow allocation if tdp enabled
  KVM/MMU: Combine flushing remote tlb in mmu_set_spte()
  KVM: vmx: skip VMWRITE of HOST_{FS,GS}_BASE when possible
  KVM: vmx: skip VMWRITE of HOST_{FS,GS}_SEL when possible
  KVM: vmx: always initialize HOST_{FS,GS}_BASE to zero during setup
  KVM: vmx: move struct host_state usage to struct loaded_vmcs
  KVM: vmx: compute need to reload FS/GS/LDT on demand
  KVM: nVMX: remove a misleading comment regarding vmcs02 fields
  KVM: vmx: rename __vmx_load_host_state() and vmx_save_host_state()
  KVM: vmx: add dedicated utility to access guest's kernel_gs_base
  KVM: vmx: track host_state.loaded using a loaded_vmcs pointer
  KVM: vmx: refactor segmentation code in vmx_save_host_state()
  kvm: nVMX: Fix fault priority for VMX operations
  kvm: nVMX: Fix fault vector for VMX operation at CPL > 0
  ...

Merge tag 'riscv-for-linus-4.19-mw0' of git://git./linux/kernel/git/palmer/riscv-linux

Pull RISC-V updates from Palmer Dabbelt:
 "This contains some major improvements to the RISC-V port, including
  the necessary interrupt controller and timer support to actually make
  it to userspace. Support for three devices has been added:

   - the ISA-mandated timers on RISC-V systems.

   - the ISA-mandated first-level interrupt controller on RISC-V
     systems, which is handled as part of our core arch code because
     it's very small and tightly tied to the ISA.

   - SiFive's platform-level interrupt controller, which talks to the
     actual devices.

  In addition to these new devices, there are a handful of cleanups all
  over the RISC-V tree:

   - build fixes for various configurations:
      * A fix to the vDSO build's makefile so it respects CFLAGS.
      * The addition of __lshrti3, a libgcc derived function necessary
        for some 32-bit configurations.
      * !SMP && PERF_EVENTS

   - Cleanups to the arch code to remove the remnants of old versions of
     the drivers that were just properly submitted.
      * Some dead code from the timer driver, most of which wasn't ever
        even compiled.
      * Cleanups of some interrupt #defines, which are now local to the
        interrupt handling code.

   - Fixes to ptrace(), which while not being sufficient to fully make
     GDB work are at least sufficient to get simple GDB tasks to work.

   - Early printk support via RISC-V's architecturally mandated SBI
     console device.

   - A fix to our early debug trap handler to ensure it's always
     aligned.

  These patches have all been through a fairly extensive review process,
  but as this enables a whole pile of functionality (ie, userspace) I'm
  confident we'll need to submit a few more patches. The only concrete
  issues I know about are the sys_riscv_flush_icache patches, but as I
  managed to screw those up on Friday I figured it'd be best to let them
  bake another week.

  This tag boots a Fedora root filesystem on QEMU's master branch for
  me, and before this morning's rebase (from 4.18-rc8 to 4.18) it booted
  on the HiFive Unleashed.

  Thanks to Christoph Hellwig and the other guys at WD for getting the
  new drivers in shape!"

* tag 'riscv-for-linus-4.19-mw0' of git://git.kernel.org/pub/scm/linux/kernel/git/palmer/riscv-linux:
  dt-bindings: interrupt-controller: SiFive Plaform Level Interrupt Controller
  dt-bindings: interrupt-controller: RISC-V local interrupt controller
  RISC-V: Fix !CONFIG_SMP compilation error
  irqchip: add a SiFive PLIC driver
  RISC-V: Add the directive for alignment of stvec's value
  clocksource: new RISC-V SBI timer driver
  RISC-V: implement low-level interrupt handling
  RISC-V: add a definition for the SIE SEIE bit
  RISC-V: remove INTERRUPT_CAUSE_* defines from asm/irq.h
  RISC-V: simplify software interrupt / IPI code
  RISC-V: remove timer leftovers
  RISC-V: Add early printk support via the SBI console
  RISC-V: Don't increment sepc after breakpoint.
  RISC-V: implement __lshrti3.
  RISC-V: Use KBUILD_CFLAGS instead of KCFLAGS when building the vDSO

Merge tag 'char-misc-4.19-rc1' of git://git./linux/kernel/git/gregkh/char-misc

Pull UIO fix from Greg KH:
 "Here is a single UIO fix that I forgot to send before 4.18-final came
  out. It reverts a UIO patch that went in the 4.18 development window
  that was causing problems.

  This patch has been in linux-next for a while with no problems, I just
  forgot to send it earlier, or as part of the larger char/misc patch
  series from yesterday, my fault"

* tag 'char-misc-4.19-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc:
  Revert "uio: use request_threaded_irq instead"

Merge branch 'for-linus' of git://git./linux/kernel/git/dtor/input

Pull input updates from Dmitry Torokhov:

 - a new driver for Rohm BU21029 touch controller

 - new bitmap APIs: bitmap_alloc, bitmap_zalloc and bitmap_free

 - updates to Atmel, eeti. pxrc and iforce drivers

 - assorted driver cleanups and fixes.

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input: (57 commits)
  MAINTAINERS: Add PhoenixRC Flight Controller Adapter
  Input: do not use WARN() in input_alloc_absinfo()
  Input: mark expected switch fall-throughs
  Input: raydium_i2c_ts - use true and false for boolean values
  Input: evdev - switch to bitmap API
  Input: gpio-keys - switch to bitmap_zalloc()
  Input: elan_i2c_smbus - cast sizeof to int for comparison
  bitmap: Add bitmap_alloc(), bitmap_zalloc() and bitmap_free()
  md: Avoid namespace collision with bitmap API
  dm: Avoid namespace collision with bitmap API
  Input: pm8941-pwrkey - add resin entry
  Input: pm8941-pwrkey - abstract register offsets and event code
  Input: iforce - reorganize joystick configuration lists
  Input: atmel_mxt_ts - move completion to after config crc is updated
  Input: atmel_mxt_ts - don't report zero pressure from T9
  Input: atmel_mxt_ts - zero terminate config firmware file
  Input: atmel_mxt_ts - refactor config update code to add context struct
  Input: atmel_mxt_ts - config CRC may start at T71
  Input: atmel_mxt_ts - remove unnecessary debug on ENOMEM
  Input: atmel_mxt_ts - remove duplicate setup of ABS_MT_PRESSURE
  ...

Merge tag 'hwlock-v4.19' of git://github.com/andersson/remoteproc

Pull hwspinlock updates from Bjorn Andersson:
 "This introduces devres helpers and an API to request a lock by name,
  then migrates the sprd SPI driver to use these"

* tag 'hwlock-v4.19' of git://github.com/andersson/remoteproc:
  hwspinlock: Fix incorrect return pointers
  spi: sprd: Change to use devm_hwspin_lock_request_specific()
  spi: sprd: Replace of_hwspin_lock_get_id() with of_hwspin_lock_get_id_byname()
  hwspinlock: Fix one comment mistake
  hwspinlock: Remove redundant config
  hwspinlock: Add devm_xxx() APIs to register/unregister one hwlock controller
  hwspinlock: Add devm_xxx() APIs to request/free hwlock
  hwspinlock: Add one new API to support getting a specific hwlock by the name

Merge tag 'rpmsg-v4.19' of git://github.com/andersson/remoteproc

Pull rpmsg updates from Bjorn Andersson:
 "This fixes a few compile and kerneldoc warnings, allows rpmsg devices
  to handle power domains, allow for labeling GLINK edges and supports
  compat for rpmsg_char"

* tag 'rpmsg-v4.19' of git://github.com/andersson/remoteproc:
  rpmsg: Add compat ioctl for rpmsg char driver
  rpmsg: glink: Store edge name for glink device
  dt-bindings: soc: qcom: Add label for GLINK bindings
  rpmsg: core: add support to power domains for devices
  rpmsg: smd: fix kerneldoc warnings
  rpmsg: glink: Fix various kerneldoc warnings.
  rpmsg: glink: correctly annotate intent members
  rpmsg: smd: Add missing include of sizes.h

Merge tag 'rproc-v4.19' of git://github.com/andersson/remoteproc

Pull remoteproc updates from Bjorn Andersson:
 "This adds support for pre-start and post-shutdown hooks for remoteproc
  subdevices, refactors the Qualcomm Hexagon support to allow reuse
  between several drivers, makes authentication in the MDT file loader
  optional, migrates a few format strings to use %pK and migrates the
  Davinci driver to use the reset framework"

* tag 'rproc-v4.19' of git://github.com/andersson/remoteproc:
  remoteproc/davinci: use the reset framework
  remoteproc/davinci: Mark error recovery as disabled
  remoteproc: st_slim: replace "%p" with "%pK"
  remoteproc: replace "%p" with "%pK"
  remoteproc: qcom: fix Q6V5_WCSS dependencies
  remoteproc: Reset table_ptr in rproc_start() failure paths
  remoteproc: qcom: q6v5-pil: fix modem hang on SDM845 after axis2 clk unvote
  remoteproc: qcom q6v5: fix modular build
  remoteproc: Introduce prepare and unprepare for subdevices
  remoteproc: rename subdev probe and remove functions
  remoteproc: Make client initialize ops in rproc_subdev
  remoteproc: Make start and stop in subdev optional
  remoteproc: Rename subdev functions to start/stop
  remoteproc: qcom: Introduce Hexagon V5 based WCSS driver
  remoteproc: qcom: q6v5-pil: Use common q6v5 helpers
  remoteproc: qcom: adsp: Use common q6v5 helpers
  remoteproc: q6v5: Extract common resource handling
  remoteproc: qcom: mdt_loader: Make the firmware authentication optional

Merge tag 'linux-watchdog-4.19-rc1' of git://linux-watchdog.org/linux-watchdog

Pull watchdog updates from Wim Van Sebroeck:

 - add MEN 16z069 IP-Core driver

 - renesas-wdt: add support for the R8A77990 wdt

 - stm32_iwdg: Add stm32mp1 support and pclk feature

 - sp805_wdt, orion_wdt, sprd_wdt: several improvements

 - imx2_wdt, stmp3xxx: switch to SPDX identifier

* tag 'linux-watchdog-4.19-rc1' of git://www.linux-watchdog.org/linux-watchdog:
  watchdog: fix dependencies of menz69_wdt.o
  watchdog: sp805: Add clock-frequency property
  watchdog: add driver for the MEN 16z069 IP-Core
  watchdog: sprd_wdt: Remove redundant dev_err call in sprd_wdt_probe()
  watchdog: stmp3xxx: Switch to SPDX identifier
  watchdog: imx2_wdt: Switch to SPDX identifier
  watchdog: sp805: set WDOG_HW_RUNNING when appropriate
  watchdog: sp805: add 'timeout-sec' DT property support
  dt-bindings: watchdog: Add optional 'timeout-sec' property for sp805
  dt-bindings: watchdog: Consolidate SP805 binding docs
  watchdog: orion_wdt: Mark watchdog as active when running at probe
  watchdog: stm32: add pclk feature for stm32mp1
  dt-bindings: watchdog: add stm32mp1 support
  dt-bindings: watchdog: renesas-wdt: Add support for the R8A77990 wdt

Merge tag 'dmaengine-4.19-rc1' of git://git.infradead.org/users/vkoul/slave-dma

Pull DMAengine updates from Vinod Koul:
 "This round brings couple of framework changes, a new driver and usual
  driver updates:

   - new managed helper for dmaengine framework registration

   - split dmaengine pause capability to pause and resume and allow
     drivers to report that individually

   - update dma_request_chan_by_mask() to handle deferred probing

   - move imx-sdma to use virt-dma

   - new driver for Actions Semi Owl family S900 controller

   - minor updates to intel, renesas, mv_xor, pl330 etc"

* tag 'dmaengine-4.19-rc1' of git://git.infradead.org/users/vkoul/slave-dma: (46 commits)
  dmaengine: Add Actions Semi Owl family S900 DMA driver
  dt-bindings: dmaengine: Add binding for Actions Semi Owl SoCs
  dmaengine: sh: rcar-dmac: Should not stop the DMAC by rcar_dmac_sync_tcr()
  dmaengine: mic_x100_dma: use the new helper to simplify the code
  dmaengine: add a new helper dmaenginem_async_device_register
  dmaengine: imx-sdma: add memcpy interface
  dmaengine: imx-sdma: add SDMA_BD_MAX_CNT to replace '0xffff'
  dmaengine: dma_request_chan_by_mask() to handle deferred probing
  dmaengine: pl330: fix irq race with terminate_all
  dmaengine: Revert "dmaengine: mv_xor_v2: enable COMPILE_TEST"
  dmaengine: mv_xor_v2: use {lower,upper}_32_bits to configure HW descriptor address
  dmaengine: mv_xor_v2: enable COMPILE_TEST
  dmaengine: mv_xor_v2: move unmap to before callback
  dmaengine: mv_xor_v2: convert callback to helper function
  dmaengine: mv_xor_v2: kill the tasklets upon exit
  dmaengine: mv_xor_v2: explicitly freeup irq
  dmaengine: sh: rcar-dmac: Add dma_pause operation
  dmaengine: sh: rcar-dmac: add a new function to clear CHCR.DE with barrier
  dmaengine: idma64: Support dmaengine_terminate_sync()
  dmaengine: hsu: Support dmaengine_terminate_sync()
  ...

Merge tag 'mmc-v4.19' of git://git./linux/kernel/git/ulfh/mmc

Pull MMC updates from Ulf Hansson:
 "Updates for MMC for v4.19.

  MMC core:
   - Add some fine-grained hooks to further support HS400 tuning
   - Improve error path for bus width setting for HS400es
   - Use a common method when checking R1 status

  MMC host:
   - renesas_sdhi: Add r8a77990 support
   - renesas_sdhi: Add eMMC HS400 mode support
   - tmio/renesas_sdhi: Improve tuning/clock management
   - tmio: Add eMMC HS400 mode support
   - sunxi: Add support for 3.3V eMMC DDR mode
   - mmci: Initial support to manage variant specific callbacks
   - sdhci: Don't try 3.3V I/O voltage if not supported
   - sdhci-pci-dwc-mshc: Add driver to support Synopsys dwc mshc SDHCI PCI
   - sdhci-of-dwcmshc: Add driver to support Synopsys DWC MSHC SDHCI
   - sdhci-msm: Add support for new version sdcc V5
   - sdhci-pci-o2micro: Add support for O2 eMMC HS200 mode
   - sdhci-pci-o2micro: Add support for O2 hardware tuning
   - sdhci-pci-o2micro: Add MSI interrupt support for O2 SD host
   - sdhci-pci: Add support for Intel ICP
   - sdhci-tegra: Prevent ACMD23 and HS200 mode on Tegra 3
   - sdhci-tegra: Fix eMMC DDR52 mode
   - sdhci-tegra: Improve clock management
   - dw_mmc-rockchip: Document compatible string for px30
   - sdhci-esdhc-imx: Add support for 3.3V eMMC DDR mode
   - sdhci-of-esdhc: Set proper DMA mask for ls104x chips
   - sdhci-of-esdhc: Improve clock management
   - sdhci-of-arasan: Add a quirk to manage unstable clocks
   - dw_mmc-exynos: Address potential external abort during system resume
   - pxamci: Add support for common MMC DT bindings
   - pxamci: Several cleanups and improvements
   - pxamci: Merge immutable branch for pxa to switch to DMA slave maps"

* tag 'mmc-v4.19' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc: (56 commits)
  mmc: core: improve reasonableness of bus width setting for HS400es
  mmc: tmio: remove unneeded variable in tmio_mmc_start_command()
  mmc: renesas_sdhi: Fix sampling clock position selecting
  mmc: tmio: Fix tuning flow
  mmc: sunxi: remove output of virtual base address
  dt-bindings: mmc: rockchip-dw-mshc: add description for px30
  mmc: renesas_sdhi: Add r8a77990 support
  mmc: sunxi: allow 3.3V DDR when DDR is available
  mmc: mmci: Add and implement a ->dma_setup() callback for qcom dml
  mmc: mmci: Initial support to manage variant specific callbacks
  mmc: tegra: Force correct divider calculation on DDR50/52
  mmc: sdhci: Add MSI interrupt support for O2 SD host
  mmc: sdhci: Add support for O2 hardware tuning
  mmc: sdhci: Export sdhci tuning function symbol
  mmc: sdhci: Change O2 Host HS200 mode clock frequency to 200MHz
  mmc: sdhci: Add support for O2 eMMC HS200 mode
  mmc: tegra: Add and use tegra_sdhci_get_max_clock()
  mmc: sdhci-esdhc-imx: fix indent
  mmc: sdhci-esdhc-imx: disable clocks before changing frequency
  mmc: tegra: prevent ACMD23 on Tegra 3
  ...

ip6_vti: simplify stats handling in vti6_xmit

Same as ip_vti, use iptunnel_xmit_stats to updates stats in tunnel xmit
code path.

Signed-off-by: Haishuang Yan <yanhaishuang@cmss.chinamobile.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

pcmcia: remove long deprecated pcmcia_request_exclusive_irq() function

This function was created as a deprecated fallback case back in 2010 by
commit eb14120f743d ("pcmcia: re-work pcmcia_request_irq()") for legacy
cases.

Actual in-kernel users haven't been around for a long while.  The last
in-kernel user was apparently removed four years ago by commit
5f5316fcd08e ("am2150: Update nmclan_cs.c to use update PCMCIA API").

Just remove it entirely.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

deprecate the '__deprecated' attribute warnings entirely and for good

We haven't had lots of deprecation warnings lately, but the rdma use of
it made them flare up again.

They are not useful.  They annoy everybody, and nobody ever does
anything about them, because it's always "somebody elses problem".  And
when people start thinking that warnings are normal, they stop looking
at them, and the real warnings that mean something go unnoticed.

If you want to get rid of a function, just get rid of it.  Convert every
user to the new world order.

And if you can't do that, then don't annoy everybody else with your
marking that says "I couldn't be bothered to fix this, so I'll just spam
everybody elses build logs with warnings about my laziness".

Make a kernelnewbies wiki page about things that could be cleaned up,
write a blog post about it, or talk to people on the mailing lists.  But
don't add warnings to the kernel build about cleanup that you think
should happen but you aren't doing yourself.

Don't.  Just don't.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Merge tag 'driver-core-4.19-rc1' of git://git./linux/kernel/git/gregkh/driver-core

Pull driver core updates from Greg KH:
 "Here are all of the driver core and related patches for 4.19-rc1.

  Nothing huge here, just a number of small cleanups and the ability to
  now stop the deferred probing after init happens.

  All of these have been in linux-next for a while with only a merge
  issue reported"

* tag 'driver-core-4.19-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core: (21 commits)
  base: core: Remove WARN_ON from link dependencies check
  drivers/base: stop new probing during shutdown
  drivers: core: Remove glue dirs from sysfs earlier
  driver core: remove unnecessary function extern declare
  sysfs.h: fix non-kernel-doc comment
  PM / Domains: Stop deferring probe at the end of initcall
  iommu: Remove IOMMU_OF_DECLARE
  iommu: Stop deferring probe at end of initcalls
  pinctrl: Support stopping deferred probe after initcalls
  dt-bindings: pinctrl: add a 'pinctrl-use-default' property
  driver core: allow stopping deferred probe after init
  driver core: add a debugfs entry to show deferred devices
  sysfs: Fix internal_create_group() for named group updates
  base: fix order of OF initialization
  linux/device.h: fix kernel-doc notation warning
  Documentation: update firmware loader fallback reference
  kobject: Replace strncpy with memcpy
  drivers: base: cacheinfo: use OF property_read_u32 instead of get_property,read_number
  kernfs: Replace strncpy with memcpy
  device: Add #define dev_fmt similar to #define pr_fmt
  ...

Merge tag 'char-misc-4.19-rc1' of git://git./linux/kernel/git/gregkh/char-misc

Pull char/misc driver updates from Greg KH:
 "Here is the bit set of char/misc drivers for 4.19-rc1

  There is a lot here, much more than normal, seems like everyone is
  writing new driver subsystems these days... Anyway, major things here
  are:

   - new FSI driver subsystem, yet-another-powerpc low-level hardware
     bus

   - gnss, finally an in-kernel GPS subsystem to try to tame all of the
     crazy out-of-tree drivers that have been floating around for years,
     combined with some really hacky userspace implementations. This is
     only for GNSS receivers, but you have to start somewhere, and this
     is great to see.

  Other than that, there are new slimbus drivers, new coresight drivers,
  new fpga drivers, and loads of DT bindings for all of these and
  existing drivers.

  All of these have been in linux-next for a while with no reported
  issues"

* tag 'char-misc-4.19-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc: (255 commits)
  android: binder: Rate-limit debug and userspace triggered err msgs
  fsi: sbefifo: Bump max command length
  fsi: scom: Fix NULL dereference
  misc: mic: SCIF Fix scif_get_new_port() error handling
  misc: cxl: changed asterisk position
  genwqe: card_base: Use true and false for boolean values
  misc: eeprom: assignment outside the if statement
  uio: potential double frees if __uio_register_device() fails
  eeprom: idt_89hpesx: clean up an error pointer vs NULL inconsistency
  misc: ti-st: Fix memory leak in the error path of probe()
  android: binder: Show extra_buffers_size in trace
  firmware: vpd: Fix section enabled flag on vpd_section_destroy
  platform: goldfish: Retire pdev_bus
  goldfish: Use dedicated macros instead of manual bit shifting
  goldfish: Add missing includes to goldfish.h
  mux: adgs1408: new driver for Analog Devices ADGS1408/1409 mux
  dt-bindings: mux: add adi,adgs1408
  Drivers: hv: vmbus: Cleanup synic memory free path
  Drivers: hv: vmbus: Remove use of slow_virt_to_phys()
  Drivers: hv: vmbus: Reset the channel callback in vmbus_onoffer_rescind()
  ...

Merge tag 'staging-4.19-rc1' of git://git./linux/kernel/git/gregkh/staging

Pull staging and IIO updates from Greg KH:
 "Here are the big staging/iio patches for 4.19-rc1.

  Lots of churn here, with tons of cleanups happening in staging
  drivers, a removal of an old crypto driver that no one was using
  (skein), and the addition of some new IIO drivers. Also added was a
  "gasket" driver from Google that needs loads of work and the erofs
  filesystem.

  Even with adding all of the new drivers and a new filesystem, we are
  only adding about 1000 lines overall to the kernel linecount, which
  shows just how much cleanup happened, and how big the unused crypto
  driver was.

  All of these have been in the linux-next tree for a while now with no
  reported issues"

* tag 'staging-4.19-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging: (903 commits)
  staging:rtl8192u: Remove unused macro definitions - Style
  staging:rtl8192u: Add spaces around '+' operator - Style
  staging:rtl8192u: Remove stale comment - Style
  staging: rtl8188eu: remove unused mp_custom_oid.h
  staging: fbtft: Add spaces around / - Style
  staging: fbtft: Erases some repetitive usage of function name - Style
  staging: fbtft: Adjust some empty-line problems - Style
  staging: fbtft: Removes one nesting level to help readability - Style
  staging: fbtft: Changes gamma table to define.
  staging: fbtft: A bit more information on dev_err.
  staging: fbtft: Fixes some alignment issues - Style
  staging: fbtft: Puts macro arguments in parenthesis to avoid precedence issues - Style
  staging: rtl8188eu: remove unused array dB_Invert_Table
  staging: rtl8188eu: remove whitespace, add missing blank line
  staging: rtl8188eu: use is_multicast_ether_addr in rtw_sta_mgt.c
  staging: rtl8188eu: remove whitespace - style
  staging: rtl8188eu: cleanup block comment - style
  staging: rtl8188eu: use is_multicast_ether_addr in rtl8188eu_xmit.c
  staging: rtl8188eu: use is_multicast_ether_addr in recv_linux.c
  staging: rtlwifi: refactor rtl_get_tcb_desc
  ...

Merge tag 'tty-4.19-rc1' of git://git./linux/kernel/git/gregkh/tty

Pull tty/serial driver updates from Greg KH:
 "Here is the big tty and serial driver pull request for 4.19-rc1.

  It's not all that big, just a number of small serial driver updates
  and fixes, along with some better vt handling for unicode characters
  for those using braille terminals.

  All of these patches have been in linux-next for a long time with no
  reported issues"

* tag 'tty-4.19-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/tty: (73 commits)
  tty: serial: 8250: Revert NXP SC16C2552 workaround
  serial: 8250_exar: Read INT0 from slave device, too
  tty: rocket: Fix possible buffer overwrite on register_PCI
  serial: 8250_dw: Add ACPI support for uart on Broadcom SoC
  serial: 8250_dw: always set baud rate in dw8250_set_termios
  dt-bindings: serial: Add binding for uartlite
  tty: serial: uartlite: Add support for suspend and resume
  tty: serial: uartlite: Add clock adaptation
  tty: serial: uartlite: Add structure for private data
  serial: sh-sci: Improve support for separate TEI and DRI interrupts
  serial: sh-sci: Remove SCIx_RZ_SCIFA_REGTYPE
  serial: sh-sci: Allow for compressed SCIF address
  serial: sh-sci: Improve interrupts description
  serial: 8250: Use cached port name directly in messages
  serial: 8250_exar: Drop unused variable in pci_xr17v35x_setup()
  vt: drop unused struct vt_struct
  vt: avoid a VLA in the unicode screen scroll function
  vt: add /dev/vcsu* to devices.txt
  vt: coherence validation code for the unicode screen buffer
  vt: selection: take screen contents from uniscr if available
  ...

Merge tag 'usb-4.19-rc1' of git://git./linux/kernel/git/gregkh/usb

Pull USB/PHY updates from Greg KH:
 "Here is the big USB and phy driver patch set for 4.19-rc1.

  Nothing huge but there was a lot of work that happened this
  development cycle:

   - lots of type-c work, with drivers graduating out of staging, and
     displayport support being added.

   - new PHY drivers

   - the normal collection of gadget driver updates and fixes

   - code churn to work on the urb handling path, using irqsave()
     everywhere in anticipation of making this codepath a lot simpler in
     the future.

   - usbserial driver fixes and reworks

   - other misc changes

  All of these have been in linux-next with no reported issues for a
  while"

* tag 'usb-4.19-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb: (159 commits)
  USB: serial: pl2303: add a new device id for ATEN
  usb: renesas_usbhs: Kconfig: convert to SPDX identifiers
  usb: dwc3: gadget: Check MaxPacketSize from descriptor
  usb: dwc2: Turn on uframe_sched on "stm32f4x9_fsotg" platforms
  usb: dwc2: Turn on uframe_sched on "amlogic" platforms
  usb: dwc2: Turn on uframe_sched on "his" platforms
  usb: dwc2: Turn on uframe_sched on "bcm" platforms
  usb: dwc2: gadget: ISOC's starting flow improvement
  usb: dwc2: Make dwc2_readl/writel functions endianness-agnostic.
  usb: dwc3: core: Enable AutoRetry feature in the controller
  usb: dwc3: Set default mode for dwc_usb31
  usb: gadget: udc: renesas_usb3: Add register of usb role switch
  usb: dwc2: replace ioread32/iowrite32_rep with dwc2_readl/writel_rep
  usb: dwc2: Modify dwc2_readl/writel functions prototype
  usb: dwc3: pci: Intel Merrifield can be host
  usb: dwc3: pci: Supply device properties via driver data
  arm64: dts: dwc3: description of incr burst type
  usb: dwc3: Enable undefined length INCR burst type
  usb: dwc3: add global soc bus configuration reg0
  usb: dwc3: Describe 'wakeup_work' field of struct dwc3_pci
  ...

Merge git://git./pub/scm/linux/kernel/git/bpf/bpf

Daniel Borkmann says:

====================
pull-request: bpf 2018-08-18

The following pull-request contains BPF updates for your *net* tree.

The main changes are:

1) Fix a BPF selftest failure in test_cgroup_storage due to rlimit
   restrictions, from Yonghong.

2) Fix a suspicious RCU rcu_dereference_check() warning triggered
   from removing a device's XDP memory allocator by using the correct
   rhashtable lookup function, from Tariq.

3) A batch of BPF sockmap and ULP fixes mainly fixing leaks and races
   as well as enforcing module aliases for ULPs. Another fix for BPF
   map redirect to make them work again with tail calls, from Daniel.

4) Fix XDP BPF samples to unload their programs upon SIGTERM, from Jesper.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

Merge git://git./pub/scm/linux/kernel/git/pablo/nf

Pablo Neira Ayuso says:

====================
Netfilter/IPVS fixes for net

The following patchset contains Netfilter/IPVS fixes for your net tree:

1) Infinite loop in IPVS when net namespace is released, from
   Tan Hu.

2) Do not show negative timeouts in ip_vs_conn by using the new
   jiffies_delta_to_msecs(), patches from Matteo Croce.

3) Set F_IFACE flag for linklocal addresses in ip6t_rpfilter,
   from Florian Westphal.

4) Fix overflow in set size allocation, from Taehee Yoo.

5) Use netlink_dump_start() from ctnetlink to fix memleak from
   the error path, again from Florian.

6) Register nfnetlink_subsys in last place, otherwise netns
   init path may lose race and see net->nft uninitialized data.
   This also reverts previous attempt to fix this by increase
   netns refcount, patches from Florian.

7) Remove conntrack entries on layer 4 protocol tracker module
   removal, from Florian.

8) Use GFP_KERNEL_ACCOUNT for xtables blob allocation, from
   Michal Hocko.

9) Get tproxy documentation in sync with existing codebase,
   from Mate Eckl.

10) Honor preset layer 3 protocol via ctx->family in the new nft_ct
    timeout infrastructure, from Harsha Sharma.

11) Let uapi nfnetlink_osf.h compile standalone with no errors,
    from Dmitry V. Levin.

12) Missing braces compilation warning in nft_tproxy, patch from
    Mate Eclk.

13) Disregard bogus check to bail out on non-anonymous sets from
    the dynamic set update extension.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

Merge tag '9p-for-4.19-2' of git://github.com/martinetd/linux

Pull 9p updates from Dominique Martinet:
 "This contains mostly fixes (6 to be backported to stable) and a few
  changes, here is the breakdown:

   - rework how fids are attributed by replacing some custom tracking in
     a list by an idr

   - for packet-based transports (virtio/rdma) validate that the packet
     length matches what the header says

   - a few race condition fixes found by syzkaller

   - missing argument check when NULL device is passed in sys_mount

   - a few virtio fixes

   - some spelling and style fixes"

* tag '9p-for-4.19-2' of git://github.com/martinetd/linux: (21 commits)
  net/9p/trans_virtio.c: add null terminal for mount tag
  9p/virtio: fix off-by-one error in sg list bounds check
  9p: fix whitespace issues
  9p: fix multiple NULL-pointer-dereferences
  fs/9p/xattr.c: catch the error of p9_client_clunk when setting xattr failed
  9p: validate PDU length
  net/9p/trans_fd.c: fix race by holding the lock
  net/9p/trans_fd.c: fix race-condition by flushing workqueue before the kfree()
  net/9p/virtio: Fix hard lockup in req_done
  net/9p/trans_virtio.c: fix some spell mistakes in comments
  9p/net: Fix zero-copy path in the 9p virtio transport
  9p: Embed wait_queue_head into p9_req_t
  9p: Replace the fidlist with an IDR
  9p: Change p9_fid_create calling convention
  9p: Fix comment on smp_wmb
  net/9p/client.c: version pointer uninitialized
  fs/9p/v9fs.c: fix spelling mistake "Uknown" -> "Unknown"
  net/9p: fix error path of p9_virtio_probe
  9p/net/protocol.c: return -ENOMEM when kmalloc() failed
  net/9p/client.c: add missing '\n' at the end of p9_debug()
  ...

Merge branch 'akpm' (patches from Andrew)

Merge updates from Andrew Morton:

 - a few misc things

 - a few Y2038 fixes

 - ntfs fixes

 - arch/sh tweaks

 - ocfs2 updates

 - most of MM

* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (111 commits)
  mm/hmm.c: remove unused variables align_start and align_end
  fs/userfaultfd.c: remove redundant pointer uwq
  mm, vmacache: hash addresses based on pmd
  mm/list_lru: introduce list_lru_shrink_walk_irq()
  mm/list_lru.c: pass struct list_lru_node* as an argument to __list_lru_walk_one()
  mm/list_lru.c: move locking from __list_lru_walk_one() to its caller
  mm/list_lru.c: use list_lru_walk_one() in list_lru_walk_node()
  mm, swap: make CONFIG_THP_SWAP depend on CONFIG_SWAP
  mm/sparse: delete old sparse_init and enable new one
  mm/sparse: add new sparse_init_nid() and sparse_init()
  mm/sparse: move buffer init/fini to the common place
  mm/sparse: use the new sparse buffer functions in non-vmemmap
  mm/sparse: abstract sparse buffer allocations
  mm/hugetlb.c: don't zero 1GiB bootmem pages
  mm, page_alloc: double zone's batchsize
  mm/oom_kill.c: document oom_lock
  mm/hugetlb: remove gigantic page support for HIGHMEM
  mm, oom: remove sleep from under oom_lock
  kernel/dma: remove unsupported gfp_mask parameter from dma_alloc_from_contiguous()
  mm/cma: remove unsupported gfp_mask parameter from cma_alloc()
  ...

mm/hmm.c: remove unused variables align_start and align_end

Variables align_start and align_end are being assigned but are never
used hence they are redundant and can be removed.

Cleans up clang warnings:
  warning: variable 'align_start' set but not used [-Wunused-but-set-variable]
  warning: variable 'align_size' set but not used [-Wunused-but-set-variable]

Link: http://lkml.kernel.org/r/20180714161124.3923-1-colin.king@canonical.com
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs/userfaultfd.c: remove redundant pointer uwq

Pointer uwq is being assigned but is never used hence it is redundant
and can be removed.

Cleans up clang warning:
  warning: variable 'uwq' set but not used [-Wunused-but-set-variable]

Link: http://lkml.kernel.org/r/20180717090802.18357-1-colin.king@canonical.com
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, vmacache: hash addresses based on pmd

When perf profiling a wide variety of different workloads, it was found
that vmacache_find() had higher than expected cost: up to 0.08% of cpu
utilization in some cases.  This was found to rival other core VM
functions such as alloc_pages_vma() with thp enabled and default
mempolicy, and the conditionals in __get_vma_policy().

VMACACHE_HASH() determines which of the four per-task_struct slots a vma
is cached for a particular address.  This currently depends on the pfn,
so pfn 5212 occupies a different vmacache slot than its neighboring pfn
5213.

vmacache_find() iterates through all four of current's vmacache slots
when looking up an address.  Hashing based on pfn, an address has
~1/VMACACHE_SIZE chance of being cached in the first vmacache slot, or
about 25%, *if* the vma is cached.

This patch hashes an address by its pmd instead of pte to optimize for
workloads with good spatial locality.  This results in a higher
probability of vmas being cached in the first slot that is checked:
normally ~70% on the same workloads instead of 25%.

[rientjes@google.com: various updates]
Link: http://lkml.kernel.org/r/alpine.DEB.2.21.1807231532290.109445@chino.kir.corp.google.com
Link: http://lkml.kernel.org/r/alpine.DEB.2.21.1807091749150.114630@chino.kir.corp.google.com
Signed-off-by: David Rientjes <rientjes@google.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/list_lru: introduce list_lru_shrink_walk_irq()

Provide list_lru_shrink_walk_irq() and let it behave like
list_lru_walk_one() except that it locks the spinlock with
spin_lock_irq().  This is used by scan_shadow_nodes() because its lock
nests within the i_pages lock which is acquired with IRQ.  This change
allows to use proper locking promitives instead hand crafted
lock_irq_disable() plus spin_lock().

There is no EXPORT_SYMBOL provided because the current user is in-kernel
only.

Add list_lru_shrink_walk_irq() which acquires the spinlock with the
proper locking primitives.

Link: http://lkml.kernel.org/r/20180716111921.5365-5-bigeasy@linutronix.de
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/list_lru.c: pass struct list_lru_node* as an argument to __list_lru_walk_one()

__list_lru_walk_one() is invoked with struct list_lru *lru, int nid as
the first two argument.  Those two are only used to retrieve struct
list_lru_node.  Since this is already done by the caller of the function
for the locking, we can pass struct list_lru_node* directly and avoid
the dance around it.

Link: http://lkml.kernel.org/r/20180716111921.5365-4-bigeasy@linutronix.de
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/list_lru.c: move locking from __list_lru_walk_one() to its caller

Move the locking inside __list_lru_walk_one() to its caller.  This is a
preparation step in order to introduce list_lru_walk_one_irq() which
does spin_lock_irq() instead of spin_lock() for the locking.

Link: http://lkml.kernel.org/r/20180716111921.5365-3-bigeasy@linutronix.de
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/list_lru.c: use list_lru_walk_one() in list_lru_walk_node()

Patch series "mm/list_lru: Add list_lru_shrink_walk_irq() and a user".

This series removes the local_irq_disable() around
list_lru_shrink_walk() (as used by mm/workingset) by adding
list_lru_shrink_walk_irq().

Vladimir Davydov preferred this over `irq' argument which I added to
struct list_lru.

The initial post (of this series) received a Reviewed-by tag by Vladimir
Davydov which I added to each patch of the series.  The series applies
on top of akpm's tree which has Kirill's shrink_slab series and does not
clash with it (akpm asked me to wait a week or so and repost it then).

I tested the code paths by triggering the OOM-killer via memory over
commit and lockdep did not complain (nor did I see any warnings).

This patch (of 4):

list_lru_walk_node() invokes __list_lru_walk_one() with -1 as the
memcg_idx parameter.  The same can be achieved by list_lru_walk_one() and
passing NULL as memcg argument which then gets converted into -1.  This is
a preparation step when the spin_lock() function is lifted to the caller
of __list_lru_walk_one().  Invoke list_lru_walk_one() instead
__list_lru_walk_one() when possible.

Link: http://lkml.kernel.org/r/20180716111921.5365-2-bigeasy@linutronix.de
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reviewed-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, swap: make CONFIG_THP_SWAP depend on CONFIG_SWAP

CONFIG_THP_SWAP should depend on CONFIG_SWAP, because it's unreasonable
to optimize swapping for THP (Transparent Huge Page) without basic
swapping support.

In original code, when CONFIG_SWAP=n and CONFIG_THP_SWAP=y,
split_swap_cluster() will not be built because it is in swapfile.c, but
it will be called in huge_memory.c.  This doesn't trigger a build error
in practice because the call site is enclosed by PageSwapCache(), which
is defined to be constant 0 when CONFIG_SWAP=n.  But this is fragile and
should be fixed.

The comments are fixed too to reflect the latest progress.

Link: http://lkml.kernel.org/r/20180713021228.439-1-ying.huang@intel.com
Fixes: 38d8b4e6bdc8 ("mm, THP, swap: delay splitting THP during swap out")
Signed-off-by: "Huang, Ying" <ying.huang@intel.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Reviewed-by: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Shaohua Li <shli@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Zi Yan <zi.yan@cs.rutgers.edu>
Cc: Daniel Jordan <daniel.m.jordan@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparse: delete old sparse_init and enable new one

Rename new_sparse_init() to sparse_init() which enables it.  Delete old
sparse_init() and all the code that became obsolete with.

[pasha.tatashin@oracle.com: remove unused sparse_mem_maps_populate_node()]
Link: http://lkml.kernel.org/r/20180716174447.14529-6-pasha.tatashin@oracle.com
Link: http://lkml.kernel.org/r/20180712203730.8703-6-pasha.tatashin@oracle.com
Signed-off-by: Pavel Tatashin <pasha.tatashin@oracle.com>
Tested-by: Michael Ellerman <mpe@ellerman.id.au> [powerpc]
Tested-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Pasha Tatashin <Pavel.Tatashin@microsoft.com>
Cc: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Daniel Jordan <daniel.m.jordan@oracle.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jan Kara <jack@suse.cz>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Souptick Joarder <jrdr.linux@gmail.com>
Cc: Steven Sistare <steven.sistare@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparse: add new sparse_init_nid() and sparse_init()

sparse_init() requires to temporary allocate two large buffers: usemap_map
and map_map.  Baoquan He has identified that these buffers are so large
that Linux is not bootable on small memory machines, such as a kdump boot.
The buffers are especially large when CONFIG_X86_5LEVEL is set, as they
are scaled to the maximum physical memory size.

Baoquan provided a fix, which reduces these sizes of these buffers, but it
is much better to get rid of them entirely.

Add a new way to initialize sparse memory: sparse_init_nid(), which only
operates within one memory node, and thus allocates memory either in large
contiguous block or allocates section by section.  This eliminates the
need for use of temporary buffers.

For simplified bisecting and review temporarly call sparse_init()
new_sparse_init(), the new interface is going to be enabled as well as old
code removed in the next patch.

Link: http://lkml.kernel.org/r/20180712203730.8703-5-pasha.tatashin@oracle.com
Signed-off-by: Pavel Tatashin <pasha.tatashin@oracle.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Tested-by: Oscar Salvador <osalvador@suse.de>
Tested-by: Michael Ellerman <mpe@ellerman.id.au> [powerpc]
Cc: Pasha Tatashin <Pavel.Tatashin@microsoft.com>
Cc: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Daniel Jordan <daniel.m.jordan@oracle.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jan Kara <jack@suse.cz>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Souptick Joarder <jrdr.linux@gmail.com>
Cc: Steven Sistare <steven.sistare@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparse: move buffer init/fini to the common place

Now that both variants of sparse memory use the same buffers to populate
memory map, we can move sparse_buffer_init()/sparse_buffer_fini() to the
common place.

Link: http://lkml.kernel.org/r/20180712203730.8703-4-pasha.tatashin@oracle.com
Signed-off-by: Pavel Tatashin <pasha.tatashin@oracle.com>
Tested-by: Michael Ellerman <mpe@ellerman.id.au> [powerpc]
Tested-by: Oscar Salvador <osalvador@suse.de>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Pasha Tatashin <Pavel.Tatashin@microsoft.com>
Cc: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Daniel Jordan <daniel.m.jordan@oracle.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jan Kara <jack@suse.cz>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Souptick Joarder <jrdr.linux@gmail.com>
Cc: Steven Sistare <steven.sistare@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparse: use the new sparse buffer functions in non-vmemmap

non-vmemmap sparse also allocated large contiguous chunk of memory, and if
fails falls back to smaller allocations.  Use the same functions to
allocate buffer as the vmemmap-sparse

Link: http://lkml.kernel.org/r/20180712203730.8703-3-pasha.tatashin@oracle.com
Signed-off-by: Pavel Tatashin <pasha.tatashin@oracle.com>
Tested-by: Michael Ellerman <mpe@ellerman.id.au> [powerpc]
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Tested-by: Oscar Salvador <osalvador@suse.de>
Cc: Pasha Tatashin <Pavel.Tatashin@microsoft.com>
Cc: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Daniel Jordan <daniel.m.jordan@oracle.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jan Kara <jack@suse.cz>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Souptick Joarder <jrdr.linux@gmail.com>
Cc: Steven Sistare <steven.sistare@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Wei Yang <richard.weiyang@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparse: abstract sparse buffer allocations

Patch series "sparse_init rewrite", v6.

In sparse_init() we allocate two large buffers to temporary hold usemap
and memmap for the whole machine.  However, we can avoid doing that if
we changed sparse_init() to operated on per-node bases instead of doing
it on the whole machine beforehand.

As shown by Baoquan
  http://lkml.kernel.org/r/20180628062857.29658-1-bhe@redhat.com

The buffers are large enough to cause machine stop to boot on small
memory systems.

Another benefit of these changes is that they also obsolete
CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER.

This patch (of 5):

When struct pages are allocated for sparse-vmemmap VA layout, we first try
to allocate one large buffer, and than if that fails allocate struct pages
for each section as we go.

The code that allocates buffer is uses global variables and is spread
across several call sites.

Cleanup the code by introducing three functions to handle the global
buffer:

sparse_buffer_init() initialize the buffer
sparse_buffer_fini() free the remaining part of the buffer
sparse_buffer_alloc() alloc from the buffer, and if buffer is empty
return NULL

Define these functions in sparse.c instead of sparse-vmemmap.c because
later we will use them for non-vmemmap sparse allocations as well.

[akpm@linux-foundation.org: use PTR_ALIGN()]
[akpm@linux-foundation.org: s/BUG_ON/WARN_ON/]
Link: http://lkml.kernel.org/r/20180712203730.8703-2-pasha.tatashin@oracle.com
Signed-off-by: Pavel Tatashin <pasha.tatashin@oracle.com>
Tested-by: Michael Ellerman <mpe@ellerman.id.au> [powerpc]
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Tested-by: Oscar Salvador <osalvador@suse.de>
Cc: Pasha Tatashin <Pavel.Tatashin@microsoft.com>
Cc: Steven Sistare <steven.sistare@oracle.com>
Cc: Daniel Jordan <daniel.m.jordan@oracle.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Souptick Joarder <jrdr.linux@gmail.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Wei Yang <richard.weiyang@gmail.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Abdul Haleem <abdhalee@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hugetlb.c: don't zero 1GiB bootmem pages

When using 1GiB pages during early boot, use the new
memblock_virt_alloc_try_nid_raw() to allocate memory without zeroing it.
Zeroing out hundreds or thousands of GiB in a single core memset() call
is very slow, and can make early boot last upwards of 20-30 minutes on
multi TiB machines.

The memory does not need to be zero'd as the hugetlb pages are always
zero'd on page fault.

Tested: Booted with ~3800 1G pages, and it booted successfully in
roughly the same amount of time as with 0, as opposed to the 25+ minutes
it would take before.

Link: http://lkml.kernel.org/r/20180711213313.92481-1-cannonmatthews@google.com
Signed-off-by: Cannon Matthews <cannonmatthews@google.com>
Acked-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Andres Lagar-Cavilla <andreslc@google.com>
Cc: Peter Feiner <pfeiner@google.com>
Cc: David Matlack <dmatlack@google.com>
Cc: Greg Thelen <gthelen@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, page_alloc: double zone's batchsize

To improve page allocator's performance for order-0 pages, each CPU has
a Per-CPU-Pageset(PCP) per zone.  Whenever an order-0 page is needed,
PCP will be checked first before asking pages from Buddy.  When PCP is
used up, a batch of pages will be fetched from Buddy to improve
performance and the size of batch can affect performance.

zone's batch size gets doubled last time by commit ba56e91c9401("mm:
page_alloc: increase size of per-cpu-pages") over ten years ago.  Since
then, CPU has envolved a lot and CPU's cache sizes also increased.

Dave Hansen is concerned the current batch size doesn't fit well with
modern hardware and suggested me to do two things: first, use a page
allocator intensive benchmark, e.g.  will-it-scale/page_fault1 to find
out how performance changes with different batch sizes on various
machines and then choose a new default batch size; second, see how this
new batch size work with other workloads.

In the first test, we saw performance gains on high-core-count systems
and little to no effect on older systems with more modest core counts.
In this phase's test data, two candidates: 63 and 127 are chosen.

In the second step, ebizzy, oltp, kbuild, pigz, netperf, vm-scalability
and more will-it-scale sub-tests are tested to see how these two
candidates work with these workloads and decides a new default according
to their results.

Most test results are flat.  will-it-scale/page_fault2 process mode has
10%-18% performance increase on 4-sockets Skylake and Broadwell.
vm-scalability/lru-file-mmap-read has 17%-47% performance increase for
4-sockets servers while for 2-sockets servers, it caused 3%-8% performance
drop.  Further analysis showed that, with a larger pcp->batch and thus
larger pcp->high(the relationship of pcp->high=6 * pcp->batch is
maintained in this patch), zone lock contention shifted to LRU add side
lock contention and that caused performance drop.  This performance drop
might be mitigated by others' work on optimizing LRU lock.

Another downside of increasing pcp->batch is, when PCP is used up and need
to fetch a batch of pages from Buddy, since batch is increased, that time
can be longer than before.  My understanding is, this doesn't affect
slowpath where direct reclaim and compaction dominates.  For fastpath,
throughput is a win(according to will-it-scale/page_fault1) but worst
latency can be larger now.

Overall, I think double the batch size from 31 to 63 is relatively safe
and provide good performance boost for high-core-count systems.

The two phase's test results are listed below(all tests are done with THP
disabled).

Phase one(will-it-scale/page_fault1) test results:

Skylake-EX: increased batch size has a good effect on zone->lock
contention, though LRU contention will rise at the same time and
limited the final performance increase.

batch   score     change   zone_contention   lru_contention   total_contention
 31   15345900    +0.00%       64%                 8%           72%
 53   17903847   +16.67%       32%                38%           70%
 63   17992886   +17.25%       24%                45%           69%
 73   18022825   +17.44%       10%                61%           71%
119   18023401   +17.45%        4%                66%           70%
127   18029012   +17.48%        3%                66%           69%
137   18036075   +17.53%        4%                66%           70%
165   18035964   +17.53%        2%                67%           69%
188   18101105   +17.95%        2%                67%           69%
223   18130951   +18.15%        2%                67%           69%
255   18118898   +18.07%        2%                67%           69%
267   18101559   +17.96%        2%                67%           69%
299   18160468   +18.34%        2%                68%           70%
320   18139845   +18.21%        2%                67%           69%
393   18160869   +18.34%        2%                68%           70%
424   18170999   +18.41%        2%                68%           70%
458   18144868   +18.24%        2%                68%           70%
467   18142366   +18.22%        2%                68%           70%
498   18154549   +18.30%        1%                68%           69%
511   18134525   +18.17%        1%                69%           70%

Broadwell-EX: similar pattern as Skylake-EX.

batch   score     change   zone_contention   lru_contention   total_contention
 31   16703983    +0.00%       67%                 7%           74%
 53   18195393    +8.93%       43%                28%           71%
 63   18288885    +9.49%       38%                33%           71%
 73   18344329    +9.82%       35%                37%           72%
119   18535529   +10.96%       24%                46%           70%
127   18513596   +10.83%       23%                48%           71%
137   18514327   +10.84%       23%                48%           71%
165   18511840   +10.82%       22%                49%           71%
188   18593478   +11.31%       17%                53%           70%
223   18601667   +11.36%       17%                52%           69%
255   18774825   +12.40%       12%                58%           70%
267   18754781   +12.28%        9%                60%           69%
299   18892265   +13.10%        7%                63%           70%
320   18873812   +12.99%        8%                62%           70%
393   18891174   +13.09%        6%                64%           70%
424   18975108   +13.60%        6%                64%           70%
458   18932364   +13.34%        8%                62%           70%
467   18960891   +13.51%        5%                65%           70%
498   18944526   +13.41%        5%                64%           69%
511   18960839   +13.51%        5%                64%           69%

Skylake-EP: although increased batch reduced zone->lock contention, but
the effect is not as good as EX: zone->lock contention is still as high as
20% with a very high batch value instead of 1% on Skylake-EX or 5% on
Broadwell-EX.  Also, total_contention actually decreased with a higher
batch but that doesn't translate to performance increase.

batch   score    change   zone_contention   lru_contention   total_contention
 31   9554867    +0.00%       66%                 3%           69%
 53   9855486    +3.15%       63%                 3%           66%
 63   9980145    +4.45%       62%                 4%           66%
 73   10092774   +5.63%       62%                 5%           67%
119   10310061   +7.90%       45%                19%           64%
127   10342019   +8.24%       42%                19%           61%
137   10358182   +8.41%       42%                21%           63%
165   10397060   +8.81%       37%                24%           61%
188   10341808   +8.24%       34%                26%           60%
223   10349135   +8.31%       31%                27%           58%
255   10327189   +8.08%       28%                29%           57%
267   10344204   +8.26%       27%                29%           56%
299   10325043   +8.06%       25%                30%           55%
320   10310325   +7.91%       25%                31%           56%
393   10293274   +7.73%       21%                31%           52%
424   10311099   +7.91%       21%                32%           53%
458   10321375   +8.02%       21%                32%           53%
467   10303881   +7.84%       21%                32%           53%
498   10332462   +8.14%       20%                33%           53%
511   10325016   +8.06%       20%                32%           52%

Broadwell-EP: zone->lock and lru lock had an agreement to make sure
performance doesn't increase and they successfully managed to keep total
contention at 70%.

batch   score    change   zone_contention   lru_contention   total_contention
 31   10121178   +0.00%       19%                50%           69%
 53   10142366   +0.21%        6%                63%           69%
 63   10117984   -0.03%       11%                58%           69%
 73   10123330   +0.02%        7%                63%           70%
119   10108791   -0.12%        2%                67%           69%
127   10166074   +0.44%        3%                66%           69%
137   10141574   +0.20%        3%                66%           69%
165   10154499   +0.33%        2%                68%           70%
188   10124921   +0.04%        2%                67%           69%
223   10137399   +0.16%        2%                67%           69%
255   10143289   +0.22%        0%                68%           68%
267   10123535   +0.02%        1%                68%           69%
299   10140952   +0.20%        0%                68%           68%
320   10163170   +0.41%        0%                68%           68%
393   10000633   -1.19%        0%                69%           69%
424   10087998   -0.33%        0%                69%           69%
458   10187116   +0.65%        0%                69%           69%
467   10146790   +0.25%        0%                69%           69%
498   10197958   +0.76%        0%                69%           69%
511   10152326   +0.31%        0%                69%           69%

Haswell-EP: similar to Broadwell-EP.

batch   score   change   zone_contention   lru_contention   total_contention
 31   10442205   +0.00%       14%                48%           62%
 53   10442255   +0.00%        5%                57%           62%
 63   10452059   +0.09%        6%                57%           63%
 73   10482349   +0.38%        5%                59%           64%
119   10454644   +0.12%        3%                60%           63%
127   10431514   -0.10%        3%                59%           62%
137   10423785   -0.18%        3%                60%           63%
165   10481216   +0.37%        2%                61%           63%
188   10448755   +0.06%        2%                61%           63%
223   10467144   +0.24%        2%                61%           63%
255   10480215   +0.36%        2%                61%           63%
267   10484279   +0.40%        2%                61%           63%
299   10466450   +0.23%        2%                61%           63%
320   10452578   +0.10%        2%                61%           63%
393   10499678   +0.55%        1%                62%           63%
424   10481454   +0.38%        1%                62%           63%
458   10473562   +0.30%        1%                62%           63%
467   10484269   +0.40%        0%                62%           62%
498   10505599   +0.61%        0%                62%           62%
511   10483395   +0.39%        0%                62%           62%

Westmere-EP: contention is pretty small so not interesting.  Note too high
a batch value could hurt performance.

batch   score   change   zone_contention   lru_contention   total_contention
 31   4831523   +0.00%        2%                 3%            5%
 53   4834086   +0.05%        2%                 4%            6%
 63   4834262   +0.06%        2%                 3%            5%
 73   4832851   +0.03%        2%                 4%            6%
119   4830534   -0.02%        1%                 3%            4%
127   4827461   -0.08%        1%                 4%            5%
137   4827459   -0.08%        1%                 3%            4%
165   4820534   -0.23%        0%                 4%            4%
188   4817947   -0.28%        0%                 3%            3%
223   4809671   -0.45%        0%                 3%            3%
255   4802463   -0.60%        0%                 4%            4%
267   4801634   -0.62%        0%                 3%            3%
299   4798047   -0.69%        0%                 3%            3%
320   4793084   -0.80%        0%                 3%            3%
393   4785877   -0.94%        0%                 3%            3%
424   4782911   -1.01%        0%                 3%            3%
458   4779346   -1.08%        0%                 3%            3%
467   4780306   -1.06%        0%                 3%            3%
498   4780589   -1.05%        0%                 3%            3%
511   4773724   -1.20%        0%                 3%            3%

Skylake-Desktop: similar to Westmere-EP, nothing interesting.

batch   score   change   zone_contention   lru_contention   total_contention
 31   3906608   +0.00%        2%                 3%            5%
 53   3940164   +0.86%        2%                 3%            5%
 63   3937289   +0.79%        2%                 3%            5%
 73   3940201   +0.86%        2%                 3%            5%
119   3933240   +0.68%        2%                 3%            5%
127   3930514   +0.61%        2%                 4%            6%
137   3938639   +0.82%        0%                 3%            3%
165   3908755   +0.05%        0%                 3%            3%
188   3905621   -0.03%        0%                 3%            3%
223   3903015   -0.09%        0%                 4%            4%
255   3889480   -0.44%        0%                 3%            3%
267   3891669   -0.38%        0%                 4%            4%
299   3898728   -0.20%        0%                 4%            4%
320   3894547   -0.31%        0%                 4%            4%
393   3875137   -0.81%        0%                 4%            4%
424   3874521   -0.82%        0%                 3%            3%
458   3880432   -0.67%        0%                 4%            4%
467   3888715   -0.46%        0%                 3%            3%
498   3888633   -0.46%        0%                 4%            4%
511   3875305   -0.80%        0%                 5%            5%

Haswell-Desktop: zone->lock is pretty low as other desktops, though lru
contention is higher than other desktops.

batch   score   change   zone_contention   lru_contention   total_contention
 31   3511158   +0.00%        2%                 5%            7%
 53   3555445   +1.26%        2%                 6%            8%
 63   3561082   +1.42%        2%                 6%            8%
 73   3547218   +1.03%        2%                 6%            8%
119   3571319   +1.71%        1%                 7%            8%
127   3549375   +1.09%        0%                 6%            6%
137   3560233   +1.40%        0%                 6%            6%
165   3555176   +1.25%        2%                 6%            8%
188   3551501   +1.15%        0%                 8%            8%
223   3531462   +0.58%        0%                 7%            7%
255   3570400   +1.69%        0%                 7%            7%
267   3532235   +0.60%        1%                 8%            9%
299   3562326   +1.46%        0%                 6%            6%
320   3553569   +1.21%        0%                 8%            8%
393   3539519   +0.81%        0%                 7%            7%
424   3549271   +1.09%        0%                 8%            8%
458   3528885   +0.50%        0%                 8%            8%
467   3526554   +0.44%        0%                 7%            7%
498   3525302   +0.40%        0%                 9%            9%
511   3527556   +0.47%        0%                 8%            8%

Sandybridge-Desktop: the 0% contention isn't accurate but caused by
dropped fractional part. Since multiple contention path's contentions
are all under 1% here, with some arithmetic operations like add, the
final deviation could be as large as 3%.

batch   score   change   zone_contention   lru_contention   total_contention
 31   1744495   +0.00%        0%                 0%            0%
 53   1755341   +0.62%        0%                 0%            0%
 63   1758469   +0.80%        0%                 0%            0%
 73   1759626   +0.87%        0%                 0%            0%
119   1770417   +1.49%        0%                 0%            0%
127   1768252   +1.36%        0%                 0%            0%
137   1767848   +1.34%        0%                 0%            0%
165   1765088   +1.18%        0%                 0%            0%
188   1766918   +1.29%        0%                 0%            0%
223   1767866   +1.34%        0%                 0%            0%
255   1768074   +1.35%        0%                 0%            0%
267   1763187   +1.07%        0%                 0%            0%
299   1765620   +1.21%        0%                 0%            0%
320   1767603   +1.32%        0%                 0%            0%
393   1764612   +1.15%        0%                 0%            0%
424   1758476   +0.80%        0%                 0%            0%
458   1758593   +0.81%        0%                 0%            0%
467   1757915   +0.77%        0%                 0%            0%
498   1753363   +0.51%        0%                 0%            0%
511   1755548   +0.63%        0%                 0%            0%

Phase two test results:
Note: all percent change is against base(batch=31).

ebizzy.throughput (higer is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1    2410037±7%     2600451±2% +7.9%     2602878 +8.0%
lkp-bdw-ex1     1493328        1489243    -0.3%     1492145 -0.1%
lkp-skl-2sp2    1329674        1345891    +1.2%     1351056 +1.6%
lkp-bdw-ep2      711511         711511     0.0%      710708 -0.1%
lkp-wsm-ep2       75750          75528    -0.3%       75441 -0.4%
lkp-skl-d01      264126         262791    -0.5%      264113 +0.0%
lkp-hsw-d01      176601         176328    -0.2%      176368 -0.1%
lkp-sb02          98937          98937    +0.0%       99030 +0.1%

kbuild.buildtime (less is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1     107.00        107.67  +0.6%        107.11  +0.1%
lkp-bdw-ex1       97.33         97.33  +0.0%         97.42  +0.1%
lkp-skl-2sp2     180.00        179.83  -0.1%        179.83  -0.1%
lkp-bdw-ep2      178.17        179.17  +0.6%        177.50  -0.4%
lkp-wsm-ep2      737.00        738.00  +0.1%        738.00  +0.1%
lkp-skl-d01      642.00        653.00  +1.7%        653.00  +1.7%
lkp-hsw-d01     1310.00       1316.00  +0.5%       1311.00  +0.1%

netperf/TCP_STREAM.Throughput_total_Mbps (higher is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1     948790        947144  -0.2%        948333 -0.0%
lkp-bdw-ex1      904224        904366  +0.0%        904926 +0.1%
lkp-skl-2sp2     239731        239607  -0.1%        239565 -0.1%
lk-bdw-ep2       365764        365933  +0.0%        365951 +0.1%
lkp-wsm-ep2       93736         93803  +0.1%         93808 +0.1%
lkp-skl-d01       77314         77303  -0.0%         77375 +0.1%
lkp-hsw-d01       58617         60387  +3.0%         60208 +2.7%
lkp-sb02          29990         30137  +0.5%         30103 +0.4%

oltp.transactions (higer is better)

machine         batch=31      batch=63             batch=127
lkp-bdw-ex1      9073276       9100377     +0.3%    9036344     -0.4%
lkp-skl-2sp2     8898717       8852054     -0.5%    8894459     -0.0%
lkp-bdw-ep2     13426155      13384654     -0.3%   13333637     -0.7%
lkp-hsw-ep2     13146314      13232784     +0.7%   13193163     +0.4%
lkp-wsm-ep2      5035355       5019348     -0.3%    5033418     -0.0%
lkp-skl-d01       418485       4413339     -0.1%    4419039     +0.0%
lkp-hsw-d01      3517817±5%    3396120±3%  -3.5%    3455138±3%  -1.8%

pigz.throughput (higer is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1    1.513e+08     1.507e+08 -0.4%      1.511e+08 -0.2%
lkp-bdw-ex1     2.060e+08     2.052e+08 -0.4%      2.044e+08 -0.8%
lkp-skl-2sp2    8.836e+08     8.845e+08 +0.1%      8.836e+08 -0.0%
lkp-bdw-ep2     8.275e+08     8.464e+08 +2.3%      8.330e+08 +0.7%
lkp-wsm-ep2     2.224e+08     2.221e+08 -0.2%      2.218e+08 -0.3%
lkp-skl-d01     1.177e+08     1.177e+08 -0.0%      1.176e+08 -0.1%
lkp-hsw-d01     1.154e+08     1.154e+08 +0.1%      1.154e+08 -0.0%
lkp-sb02        0.633e+08     0.633e+08 +0.1%      0.633e+08 +0.0%

will-it-scale.malloc1.processes (higher is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1      620181       620484 +0.0%         620240 +0.0%
lkp-bdw-ex1      1403610      1401201 -0.2%        1417900 +1.0%
lkp-skl-2sp2     1288097      1284145 -0.3%        1283907 -0.3%
lkp-bdw-ep2      1427879      1427675 -0.0%        1428266 +0.0%
lkp-hsw-ep2      1362546      1353965 -0.6%        1354759 -0.6%
lkp-wsm-ep2      2099657      2107576 +0.4%        2100226 +0.0%
lkp-skl-d01      1476835      1476358 -0.0%        1474487 -0.2%
lkp-hsw-d01      1308810      1303429 -0.4%        1301299 -0.6%
lkp-sb02          589286       589284 -0.0%         588101 -0.2%

will-it-scale.malloc1.threads (higher is better)
machine         batch=31      batch=63             batch=127
lkp-skl-4sp1     21289         21125     -0.8%      21241     -0.2%
lkp-bdw-ex1      28114         28089     -0.1%      28007     -0.4%
lkp-skl-2sp2     91866         91946     +0.1%      92723     +0.9%
lkp-bdw-ep2      37637         37501     -0.4%      37317     -0.9%
lkp-hsw-ep2      43673         43590     -0.2%      43754     +0.2%
lkp-wsm-ep2      28577         28298     -1.0%      28545     -0.1%
lkp-skl-d01     175277        173343     -1.1%     173082     -1.3%
lkp-hsw-d01     130303        129566     -0.6%     129250     -0.8%
lkp-sb02        113742±3%     116911     +2.8%     116417±3%  +2.4%

will-it-scale.malloc2.processes (higer is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1    1.206e+09     1.206e+09 -0.0%      1.206e+09 +0.0%
lkp-bdw-ex1     1.319e+09     1.319e+09 -0.0%      1.319e+09 +0.0%
lkp-skl-2sp2    8.000e+08     8.021e+08 +0.3%      7.995e+08 -0.1%
lkp-bdw-ep2     6.582e+08     6.634e+08 +0.8%      6.513e+08 -1.1%
lkp-hsw-ep2     6.671e+08     6.669e+08 -0.0%      6.665e+08 -0.1%
lkp-wsm-ep2     1.805e+08     1.806e+08 +0.0%      1.804e+08 -0.1%
lkp-skl-d01     1.611e+08     1.611e+08 -0.0%      1.610e+08 -0.0%
lkp-hsw-d01     1.333e+08     1.332e+08 -0.0%      1.332e+08 -0.0%
lkp-sb02         82485104      82478206 -0.0%       82473546 -0.0%

will-it-scale.malloc2.threads (higer is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1    1.574e+09     1.574e+09 -0.0%      1.574e+09 -0.0%
lkp-bdw-ex1     1.737e+09     1.737e+09 +0.0%      1.737e+09 -0.0%
lkp-skl-2sp2    9.161e+08     9.162e+08 +0.0%      9.181e+08 +0.2%
lkp-bdw-ep2     7.856e+08     8.015e+08 +2.0%      8.113e+08 +3.3%
lkp-hsw-ep2     6.908e+08     6.904e+08 -0.1%      6.907e+08 -0.0%
lkp-wsm-ep2     2.409e+08     2.409e+08 +0.0%      2.409e+08 -0.0%
lkp-skl-d01     1.199e+08     1.199e+08 -0.0%      1.199e+08 -0.0%
lkp-hsw-d01     1.029e+08     1.029e+08 -0.0%      1.029e+08 +0.0%
lkp-sb02         68081213      68061423 -0.0%       68076037 -0.0%

will-it-scale.page_fault2.processes (higer is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1    14509125±4%   16472364 +13.5%       17123117 +18.0%
lkp-bdw-ex1     14736381      16196588  +9.9%       16364011 +11.0%
lkp-skl-2sp2     6354925       6435444  +1.3%        6436644  +1.3%
lkp-bdw-ep2      8749584       8834422  +1.0%        8827179  +0.9%
lkp-hsw-ep2      8762591       8845920  +1.0%        8825697  +0.7%
lkp-wsm-ep2      3036083       3030428  -0.2%        3021741  -0.5%
lkp-skl-d01      2307834       2304731  -0.1%        2286142  -0.9%
lkp-hsw-d01      1806237       1800786  -0.3%        1795943  -0.6%
lkp-sb02          842616        837844  -0.6%         833921  -1.0%

will-it-scale.page_fault2.threads

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1     1623294       1615132±2% -0.5%     1656777    +2.1%
lkp-bdw-ex1      1995714       2025948    +1.5%     2113753±3% +5.9%
lkp-skl-2sp2     2346708       2415591    +2.9%     2416919    +3.0%
lkp-bdw-ep2      2342564       2344882    +0.1%     2300206    -1.8%
lkp-hsw-ep2      1820658       1831681    +0.6%     1844057    +1.3%
lkp-wsm-ep2      1725482       1733774    +0.5%     1740517    +0.9%
lkp-skl-d01      1832833       1823628    -0.5%     1806489    -1.4%
lkp-hsw-d01      1427913       1427287    -0.0%     1420226    -0.5%
lkp-sb02          750626        748615    -0.3%      746621    -0.5%

will-it-scale.page_fault3.processes (higher is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1    24382726      24400317 +0.1%       24668774 +1.2%
lkp-bdw-ex1     35399750      35683124 +0.8%       35829492 +1.2%
lkp-skl-2sp2    28136820      28068248 -0.2%       28147989 +0.0%
lkp-bdw-ep2     37269077      37459490 +0.5%       37373073 +0.3%
lkp-hsw-ep2     36224967      36114085 -0.3%       36104908 -0.3%
lkp-wsm-ep2     16820457      16911005 +0.5%       16968596 +0.9%
lkp-skl-d01      7721138       7725904 +0.1%        7756740 +0.5%
lkp-hsw-d01      7611979       7650928 +0.5%        7651323 +0.5%
lkp-sb02         3781546       3796502 +0.4%        3796827 +0.4%

will-it-scale.page_fault3.threads (higer is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1     1865820±3%   1900917±2%  +1.9%     1826245±4%  -2.1%
lkp-bdw-ex1      3094060      3148326     +1.8%     3150036     +1.8%
lkp-skl-2sp2     3952940      3953898     +0.0%     3989360     +0.9%
lkp-bdw-ep2      3420373±3%   3643964     +6.5%     3644910±5%  +6.6%
lkp-hsw-ep2      2609635±2%   2582310±3%  -1.0%     2780459     +6.5%
lkp-wsm-ep2      4395001      4417196     +0.5%     4432499     +0.9%
lkp-skl-d01      5363977      5400003     +0.7%     5411370     +0.9%
lkp-hsw-d01      5274131      5311294     +0.7%     5319359     +0.9%
lkp-sb02         2917314      2913004     -0.1%     2935286     +0.6%

will-it-scale.read1.processes (higer is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1    73762279±14%  69322519±10% -6.0%    69349855±13%  -6.0% (result unstable)
lkp-bdw-ex1     1.701e+08     1.704e+08    +0.1%    1.705e+08     +0.2%
lkp-skl-2sp2    63111570      63113953     +0.0%    63836573      +1.1%
lkp-bdw-ep2     79247409      79424610     +0.2%    78012656      -1.6%
lkp-hsw-ep2     67677026      68308800     +0.9%    67539106      -0.2%
lkp-wsm-ep2     13339630      13939817     +4.5%    13766865      +3.2%
lkp-skl-d01     10969487      10972650     +0.0%    no data
lkp-hsw-d01     9857342±2%    10080592±2%  +2.3%    10131560      +2.8%
lkp-sb02        5189076        5197473     +0.2%    5163253       -0.5%

will-it-scale.read1.threads (higher is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1    62468045±12%  73666726±7% +17.9%    79553123±12% +27.4% (result unstable)
lkp-bdw-ex1     1.62e+08      1.624e+08    +0.3%    1.614e+08     -0.3%
lkp-skl-2sp2    58319780      59181032     +1.5%    59821353      +2.6%
lkp-bdw-ep2     74057992      75698171     +2.2%    74990869      +1.3%
lkp-hsw-ep2     63672959      63639652     -0.1%    64387051      +1.1%
lkp-wsm-ep2     13489943      13526058     +0.3%    13259032      -1.7%
lkp-skl-d01     10297906      10338796     +0.4%    10407328      +1.1%
lkp-hsw-d01      9636721       9667376     +0.3%     9341147      -3.1%
lkp-sb02         4801938       4804496     +0.1%     4802290      +0.0%

will-it-scale.write1.processes (higer is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1    1.111e+08     1.104e+08±2%  -0.7%   1.122e+08±2%  +1.0%
lkp-bdw-ex1     1.392e+08     1.399e+08     +0.5%   1.397e+08     +0.4%
lkp-skl-2sp2     59369233      58994841     -0.6%    58715168     -1.1%
lkp-bdw-ep2      61820979      CPU throttle          63593123     +2.9%
lkp-hsw-ep2      57897587      57435605     -0.8%    56347450     -2.7%
lkp-wsm-ep2       7814203       7918017±2%  +1.3%     7669068     -1.9%
lkp-skl-d01       8886557       8971422     +1.0%     8818366     -0.8%
lkp-hsw-d01       9171001±5%    9189915     +0.2%     9483909     +3.4%
lkp-sb02          4475406       4475294     -0.0%     4501756     +0.6%

will-it-scale.write1.threads (higer is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1    1.058e+08     1.055e+08±2%  -0.2%   1.065e+08  +0.7%
lkp-bdw-ex1     1.316e+08     1.300e+08     -1.2%   1.308e+08  -0.6%
lkp-skl-2sp2     54492421      56086678     +2.9%    55975657  +2.7%
lkp-bdw-ep2      59360449      59003957     -0.6%    58101262  -2.1%
lkp-hsw-ep2      53346346±2%   52530876     -1.5%    52902487  -0.8%
lkp-wsm-ep2       7774006       7800092±2%  +0.3%     7558833  -2.8%
lkp-skl-d01       8346174       8235695     -1.3%     no data
lkp-hsw-d01       8636244       8655731     +0.2%     8658868  +0.3%
lkp-sb02          4181820       4204107     +0.5%     4182992  +0.0%

vm-scalability.anon-r-rand.throughput (higher is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1    11933873±3%   12356544±2%  +3.5%   12188624     +2.1%
lkp-bdw-ex1      7114424±2%    7330949±2%  +3.0%    7392419     +3.9%
lkp-skl-2sp2     6773277±5%    6492332±8%  -4.1%    6543962     -3.4%
lkp-bdw-ep2      7133846±4%    7233508     +1.4%    7013518±3%  -1.7%
lkp-hsw-ep2      4576626       4527098     -1.1%    4551679     -0.5%
lkp-wsm-ep2      2583599       2592492     +0.3%    2588039     +0.2%
lkp-hsw-d01       998199±2%    1028311     +3.0%    1006460±2%  +0.8%
lkp-sb02          570572        567854     -0.5%     568449     -0.4%

vm-scalability.anon-r-rand-mt.throughput (higher is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1     1789419       1787830     -0.1%    1788208     -0.1%
lkp-bdw-ex1      3492595±2%    3554966±2%  +1.8%    3558835±3%  +1.9%
lkp-skl-2sp2     3856238±2%    3975403±4%  +3.1%    3994600     +3.6%
lkp-bdw-ep2      3726963±11%   3809292±6%  +2.2%    3871924±4%  +3.9%
lkp-hsw-ep2      2131760±3%    2033578±4%  -4.6%    2130727±6%  -0.0%
lkp-wsm-ep2      2369731       2368384     -0.1%    2370252     +0.0%
lkp-skl-d01      1207128       1206220     -0.1%    1205801     -0.1%
lkp-hsw-d01       964317        992329±2%  +2.9%     992099±2%  +2.9%
lkp-sb02          567137        567346     +0.0%     566144     -0.2%

vm-scalability.lru-file-mmap-read.throughput (higher is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1    19560469±6%   23018999     +17.7%   23418800     +19.7%
lkp-bdw-ex1     17769135±14%  26141676±3%  +47.1%   26284723±5%  +47.9%
lkp-skl-2sp2    14056512      13578884      -3.4%   13146214      -6.5%
lkp-bdw-ep2     15336542      14737654      -3.9%   14088159      -8.1%
lkp-hsw-ep2     16275498      15756296      -3.2%   15018090      -7.7%
lkp-wsm-ep2     11272160      11237231      -0.3%   11310047      +0.3%
lkp-skl-d01      7322119       7324569      +0.0%    7184148      -1.9%
lkp-hsw-d01      6449234       6404542      -0.7%    6356141      -1.4%
lkp-sb02         3517943       3520668      +0.1%    3527309      +0.3%

vm-scalability.lru-file-mmap-read-rand.throughput (higher is better)

machine         batch=31      batch=63             batch=127
lkp-skl-4sp1     1689052       1697553  +0.5%       1698726  +0.6%
lkp-bdw-ex1      1675246       1699764  +1.5%       1712226  +2.2%
lkp-skl-2sp2     1800533       1799749  -0.0%       1800581  +0.0%
lkp-bdw-ep2      1807422       1807758  +0.0%       1804932  -0.1%
lkp-hsw-ep2      1809807       1808781  -0.1%       1807811  -0.1%
lkp-wsm-ep2      1800198       1802434  +0.1%       1801236  +0.1%
lkp-skl-d01       696689        695537  -0.2%        694106  -0.4%
lkp-hsw-d01       698364        698666  +0.0%        696686  -0.2%
lkp-sb02          258939        258787  -0.1%        258199  -0.3%

Link: http://lkml.kernel.org/r/20180711055855.29072-1-aaron.lu@intel.com
Signed-off-by: Aaron Lu <aaron.lu@intel.com>
Suggested-by: Dave Hansen <dave.hansen@intel.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Huang Ying <ying.huang@intel.com>
Cc: Kemi Wang <kemi.wang@intel.com>
Cc: Tim Chen <tim.c.chen@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/oom_kill.c: document oom_lock

Add comments describing oom_lock's scope.

Requested-by: David Rientjes <rientjes@google.com>
Link: http://lkml.kernel.org/r/20180711120121.25635-1-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/hugetlb: remove gigantic page support for HIGHMEM

This reverts ee8f248d266e ("hugetlb: add phys addr to struct
huge_bootmem_page").

At one time powerpc used this field and supporting code.  However that
was removed with commit 79cc38ded1e1 ("powerpc/mm/hugetlb: Add support
for reserving gigantic huge pages via kernel command line").

There are no users of this field and supporting code, so remove it.

Link: http://lkml.kernel.org/r/20180711195913.1294-1-mike.kravetz@oracle.com
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: "Aneesh Kumar K . V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Cannon Matthews <cannonmatthews@google.com>
Cc: Becky Bruce <beckyb@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, oom: remove sleep from under oom_lock

Tetsuo has pointed out that since 27ae357fa82b ("mm, oom: fix concurrent
munlock and oom reaper unmap, v3") we have a strong synchronization
between the oom_killer and victim's exiting because both have to take
the oom_lock.  Therefore the original heuristic to sleep for a short
time in out_of_memory doesn't serve the original purpose.

Moreover Tetsuo has noticed that the short sleep can be more harmful
than actually useful.  Hammering the system with many processes can lead
to a starvation when the task holding the oom_lock can block for a long
time (minutes) and block any further progress because the oom_reaper
depends on the oom_lock as well.

Drop the short sleep from out_of_memory when we hold the lock.  Keep the
sleep when the trylock fails to throttle the concurrent OOM paths a bit.
This should be solved in a more reasonable way (e.g.  sleep proportional
to the time spent in the active reclaiming etc.) but this is much more
complex thing to achieve.  This is a quick fixup to remove a stale code.

Link: http://lkml.kernel.org/r/20180709074706.30635-1-mhocko@kernel.org
Signed-off-by: Michal Hocko <mhocko@suse.com>
Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

kernel/dma: remove unsupported gfp_mask parameter from dma_alloc_from_contiguous()

The CMA memory allocator doesn't support standard gfp flags for memory
allocation, so there is no point having it as a parameter for
dma_alloc_from_contiguous() function.  Replace it by a boolean no_warn
argument, which covers all the underlaying cma_alloc() function
supports.

This will help to avoid giving false feeling that this function supports
standard gfp flags and callers can pass __GFP_ZERO to get zeroed buffer,
what has already been an issue: see commit dd65a941f6ba ("arm64:
dma-mapping: clear buffers allocated with FORCE_CONTIGUOUS flag").

Link: http://lkml.kernel.org/r/20180709122020eucas1p21a71b092975cb4a3b9954ffc63f699d1~-sqUFoa-h2939329393eucas1p2Y@eucas1p2.samsung.com
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Michał Nazarewicz <mina86@mina86.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Laura Abbott <labbott@redhat.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Joonsoo Kim <js1304@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/cma: remove unsupported gfp_mask parameter from cma_alloc()

cma_alloc() doesn't really support gfp flags other than __GFP_NOWARN, so
convert gfp_mask parameter to boolean no_warn parameter.

This will help to avoid giving false feeling that this function supports
standard gfp flags and callers can pass __GFP_ZERO to get zeroed buffer,
what has already been an issue: see commit dd65a941f6ba ("arm64:
dma-mapping: clear buffers allocated with FORCE_CONTIGUOUS flag").

Link: http://lkml.kernel.org/r/20180709122019eucas1p2340da484acfcc932537e6014f4fd2c29~-sqTPJKij2939229392eucas1p2j@eucas1p2.samsung.com
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Michał Nazarewicz <mina86@mina86.com>
Acked-by: Laura Abbott <labbott@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Cc: Joonsoo Kim <js1304@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Revert "mm: always flush VMA ranges affected by zap_page_range"

There was a bug in Linux that could cause madvise (and mprotect?) system
calls to return to userspace without the TLB having been flushed for all
the pages involved.

This could happen when multiple threads of a process made simultaneous
madvise and/or mprotect calls.

This was noticed in the summer of 2017, at which time two solutions
were created:

  56236a59556c ("mm: refactor TLB gathering API")
  99baac21e458 ("mm: fix MADV_[FREE|DONTNEED] TLB flush miss problem")
and
  4647706ebeee ("mm: always flush VMA ranges affected by zap_page_range")

We need only one of these solutions, and the former appears to be a
little more efficient than the latter, so revert that one.

This reverts 4647706ebeee6e50 ("mm: always flush VMA ranges affected by
zap_page_range")

Link: http://lkml.kernel.org/r/20180706131019.51e3a5f0@imladris.surriel.com
Signed-off-by: Rik van Riel <riel@surriel.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: "Aneesh Kumar K.V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Nadav Amit <nadav.amit@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparse: optimize memmap allocation during sparse_init()

In sparse_init(), two temporary pointer arrays, usemap_map and map_map
are allocated with the size of NR_MEM_SECTIONS.  They are used to store
each memory section's usemap and mem map if marked as present.  With the
help of these two arrays, continuous memory chunk is allocated for
usemap and memmap for memory sections on one node.  This avoids too many
memory fragmentations.  Like below diagram, '1' indicates the present
memory section, '0' means absent one.  The number 'n' could be much
smaller than NR_MEM_SECTIONS on most of systems.

  |1|1|1|1|0|0|0|0|1|1|0|0|...|1|0||1|0|...|1||0|1|...|0|
  -------------------------------------------------------
   0 1 2 3         4 5         i   i+1     n-1   n

If we fail to populate the page tables to map one section's memmap, its
->section_mem_map will be cleared finally to indicate that it's not
present.  After use, these two arrays will be released at the end of
sparse_init().

In 4-level paging mode, each array costs 4M which can be ignorable.
While in 5-level paging, they costs 256M each, 512M altogether.  Kdump
kernel Usually only reserves very few memory, e.g 256M.  So, even thouth
they are temporarily allocated, still not acceptable.

In fact, there's no need to allocate them with the size of
NR_MEM_SECTIONS.  Since the ->section_mem_map clearing has been deferred
to the last, the number of present memory sections are kept the same
during sparse_init() until we finally clear out the memory section's
->section_mem_map if its usemap or memmap is not correctly handled.
Thus in the middle whenever for_each_present_section_nr() loop is taken,
the i-th present memory section is always the same one.

Here only allocate usemap_map and map_map with the size of
'nr_present_sections'.  For the i-th present memory section, install its
usemap and memmap to usemap_map[i] and mam_map[i] during allocation.
Then in the last for_each_present_section_nr() loop which clears the
failed memory section's ->section_mem_map, fetch usemap and memmap from
usemap_map[] and map_map[] array and set them into mem_section[]
accordingly.

[akpm@linux-foundation.org: coding-style fixes]
Link: http://lkml.kernel.org/r/20180628062857.29658-5-bhe@redhat.com
Signed-off-by: Baoquan He <bhe@redhat.com>
Reviewed-by: Pavel Tatashin <pasha.tatashin@oracle.com>
Cc: Pasha Tatashin <Pavel.Tatashin@microsoft.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Oscar Salvador <osalvador@techadventures.net>
Cc: Pankaj Gupta <pagupta@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparse.c: add a new parameter 'data_unit_size' for alloc_usemap_and_memmap

It's used to pass the size of map data unit into
alloc_usemap_and_memmap, and is preparation for next patch.

Link: http://lkml.kernel.org/r/20180228032657.32385-4-bhe@redhat.com
Signed-off-by: Baoquan He <bhe@redhat.com>
Reviewed-by: Pavel Tatashin <pasha.tatashin@oracle.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Pasha Tatashin <Pavel.Tatashin@microsoft.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Pankaj Gupta <pagupta@redhat.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparsemem.c: defer the ms->section_mem_map clearing

In sparse_init(), if CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER=y, system
will allocate one continuous memory chunk for mem maps on one node and
populate the relevant page tables to map memory section one by one.  If
fail to populate for a certain mem section, print warning and its
->section_mem_map will be cleared to cancel the marking of being
present.  Like this, the number of mem sections marked as present could
become less during sparse_init() execution.

Here just defer the ms->section_mem_map clearing if failed to populate
its page tables until the last for_each_present_section_nr() loop.  This
is in preparation for later optimizing the mem map allocation.

[akpm@linux-foundation.org: remove now-unused local `ms', per Oscar]
Link: http://lkml.kernel.org/r/20180228032657.32385-3-bhe@redhat.com
Signed-off-by: Baoquan He <bhe@redhat.com>
Acked-by: Dave Hansen <dave.hansen@intel.com>
Reviewed-by: Pavel Tatashin <pasha.tatashin@oracle.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Pasha Tatashin <Pavel.Tatashin@microsoft.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Pankaj Gupta <pagupta@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/sparse.c: add a static variable nr_present_sections

Patch series "mm/sparse: Optimize memmap allocation during
sparse_init()", v6.

In sparse_init(), two temporary pointer arrays, usemap_map and map_map
are allocated with the size of NR_MEM_SECTIONS.  They are used to store
each memory section's usemap and mem map if marked as present.  In
5-level paging mode, this will cost 512M memory though they will be
released at the end of sparse_init().  System with few memory, like
kdump kernel which usually only has about 256M, will fail to boot
because of allocation failure if CONFIG_X86_5LEVEL=y.

In this patchset, optimize the memmap allocation code to only use
usemap_map and map_map with the size of nr_present_sections.  This makes
kdump kernel boot up with normal crashkernel='' setting when
CONFIG_X86_5LEVEL=y.

This patch (of 5):

nr_present_sections is used to record how many memory sections are
marked as present during system boot up, and will be used in the later
patch.

Link: http://lkml.kernel.org/r/20180228032657.32385-2-bhe@redhat.com
Signed-off-by: Baoquan He <bhe@redhat.com>
Acked-by: Dave Hansen <dave.hansen@intel.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Pavel Tatashin <pasha.tatashin@oracle.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Pasha Tatashin <Pavel.Tatashin@microsoft.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Pankaj Gupta <pagupta@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: use special value SHRINKER_REGISTERING instead of list_empty() check

The patch introduces a special value SHRINKER_REGISTERING to use instead
of list_empty() to differ a registering shrinker from unregistered
shrinker.  Why we need that at all?

Shrinker registration is split in two parts.  The first one is
prealloc_shrinker(), which allocates shrinker memory and reserves ID in
shrinker_idr.  This function can fail.  The second is
register_shrinker_prepared(), and it finalizes the registration.  This
function actually makes shrinker available to be used from
shrink_slab(), and it can't fail.

One shrinker may be based on more then one LRU lists.  So, we never
clear the bit in memcg shrinker maps, when (one of) corresponding LRU
list becomes empty, since other LRU lists may be not empty.  See
superblock shrinker for example: it is based on two LRU lists:
s_inode_lru and s_dentry_lru.  We do not want to clear shrinker bit,
when there are no inodes in s_inode_lru, as s_dentry_lru may contain
dentries.

Instead of that, we use special algorithm to detect shrinkers having no
elements at all its LRU lists, and this is made in shrink_slab_memcg().
See the comment in this function for the details.

Also, in shrink_slab_memcg() we clear shrinker bit in the map, when we
meet unregistered shrinker (bit is set, while there is no a shrinker in
IDR).  Otherwise, we would have done that at the moment of shrinker
unregistration for all memcgs (and this looks worse, since iteration
over all memcg may take much time).  Also this would have imposed
restrictions on shrinker unregistration order for its users: they would
have had to guarantee, there are no new elements after
unregister_shrinker() (otherwise, a new added element would have set a
bit).

So, if we meet a set bit in map and no shrinker in IDR when we're
iterating over the map in shrink_slab_memcg(), this means the
corresponding shrinker is unregistered, and we must clear the bit.

Another case is shrinker registration.  We want two things there:

1) do_shrink_slab() can be called only for completely registered
   shrinkers;

2) shrinker internal lists may be populated in any order with
   register_shrinker_prepared() (let's talk on the example with sb).  Both
   of:

  a)list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru); [cpu0]
    memcg_set_shrinker_bit();                               [cpu0]
    ...
    register_shrinker_prepared();                           [cpu1]

  and

  b)register_shrinker_prepared();                           [cpu0]
    ...
    list_lru_add(&inode->i_sb->s_inode_lru, &inode->i_lru); [cpu1]
    memcg_set_shrinker_bit();                               [cpu1]

   are legitimate.  We don't want to impose restriction here and to
   force people to use only (b) variant.  We don't want to force people to
   care, there is no elements in LRU lists before the shrinker is
   completely registered.  Internal users of LRU lists and shrinker code
   are two different subsystems, and they have to be closed in themselves
   each other.

In (a) case we have the bit set before shrinker is completely
registered.  We don't want do_shrink_slab() is called at this moment, so
we have to detect such the registering shrinkers.

Before this patch list_empty() (shrinker is not linked to the list)
check was used for that.  So, in (a) there could be a bit set, but we
don't call do_shrink_slab() unless shrinker is linked to the list.  It's
just an indicator, I just overloaded linking to the list.

This was not the best solution, since it's better not to touch the
shrinker memory from shrink_slab_memcg() before it's completely
registered (this also will be useful in the future to make shrink_slab()
completely lockless).

So, this patch introduces better way to detect registering shrinker,
which allows not to dereference shrinker memory.  It's just a ~0UL
value, which we insert into the IDR during ID allocation.  After
shrinker is ready to be used, we insert actual shrinker pointer in the
IDR, and it becomes available to shrink_slab_memcg().

We can't use NULL instead of this new value for this purpose as:
shrink_slab_memcg() already uses NULL to detect unregistered shrinkers,
and we don't want the function sees NULL and clears the bit, otherwise
(a) won't work.

This is the only thing the patch makes: the better way to detect
registering shrinker.  Nothing else this patch makes.

Also this gives a better assembler, but it's minor side of the patch:

Before:
  callq  <idr_find>
  mov    %rax,%r15
  test   %rax,%rax
  je     <shrink_slab_memcg+0x1d5>
  mov    0x20(%rax),%rax
  lea    0x20(%r15),%rdx
  cmp    %rax,%rdx
  je     <shrink_slab_memcg+0xbd>
  mov    0x8(%rsp),%edx
  mov    %r15,%rsi
  lea    0x10(%rsp),%rdi
  callq  <do_shrink_slab>

After:
  callq  <idr_find>
  mov    %rax,%r15
  lea    -0x1(%rax),%rax
  cmp    $0xfffffffffffffffd,%rax
  ja     <shrink_slab_memcg+0x1cd>
  mov    0x8(%rsp),%edx
  mov    %r15,%rsi
  lea    0x10(%rsp),%rdi
  callq  ffffffff810cefd0 <do_shrink_slab>

[ktkhai@virtuozzo.com: add #ifdef CONFIG_MEMCG_KMEM around idr_replace()]
Link: http://lkml.kernel.org/r/758b8fec-7573-47eb-b26a-7b2847ae7b8c@virtuozzo.com
Link: http://lkml.kernel.org/r/153355467546.11522.4518015068123480218.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Josef Bacik <jbacik@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan.c: move check for SHRINKER_NUMA_AWARE to do_shrink_slab()

In case of shrink_slab_memcg() we do not zero nid, when shrinker is not
numa-aware.  This is not a real problem, since currently all memcg-aware
shrinkers are numa-aware too (we have two: super_block shrinker and
workingset shrinker), but something may change in the future.

Link: http://lkml.kernel.org/r/153320759911.18959.8842396230157677671.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Josef Bacik <jbacik@fb.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan.c: clear shrinker bit if there are no objects related to memcg

To avoid further unneed calls of do_shrink_slab() for shrinkers, which
already do not have any charged objects in a memcg, their bits have to
be cleared.

This patch introduces a lockless mechanism to do that without races
without parallel list lru add.  After do_shrink_slab() returns
SHRINK_EMPTY the first time, we clear the bit and call it once again.
Then we restore the bit, if the new return value is different.

Note, that single smp_mb__after_atomic() in shrink_slab_memcg() covers
two situations:

1)list_lru_add()     shrink_slab_memcg
    list_add_tail()    for_each_set_bit() <--- read bit
                         do_shrink_slab() <--- missed list update (no barrier)
    <MB>                 <MB>
    set_bit()            do_shrink_slab() <--- seen list update

This situation, when the first do_shrink_slab() sees set bit, but it
doesn't see list update (i.e., race with the first element queueing), is
rare.  So we don't add <MB> before the first call of do_shrink_slab()
instead of this to do not slow down generic case.  Also, it's need the
second call as seen in below in (2).

2)list_lru_add()      shrink_slab_memcg()
    list_add_tail()     ...
    set_bit()           ...
  ...                   for_each_set_bit()
  do_shrink_slab()        do_shrink_slab()
    clear_bit()           ...
  ...                     ...
  list_lru_add()          ...
    list_add_tail()       clear_bit()
    <MB>                  <MB>
    set_bit()             do_shrink_slab()

The barriers guarantee that the second do_shrink_slab() in the right
side task sees list update if really cleared the bit.  This case is
drawn in the code comment.

[Results/performance of the patchset]

After the whole patchset applied the below test shows signify increase
of performance:

  $echo 1 > /sys/fs/cgroup/memory/memory.use_hierarchy
  $mkdir /sys/fs/cgroup/memory/ct
  $echo 4000M > /sys/fs/cgroup/memory/ct/memory.kmem.limit_in_bytes
      $for i in `seq 0 4000`; do mkdir /sys/fs/cgroup/memory/ct/$i;
    echo $$ > /sys/fs/cgroup/memory/ct/$i/cgroup.procs;
    mkdir -p s/$i; mount -t tmpfs $i s/$i;
    touch s/$i/file; done

Then, 5 sequential calls of drop caches:

  $time echo 3 > /proc/sys/vm/drop_caches

1)Before:
  0.00user 13.78system 0:13.78elapsed 99%CPU
  0.00user 5.59system 0:05.60elapsed 99%CPU
  0.00user 5.48system 0:05.48elapsed 99%CPU
  0.00user 8.35system 0:08.35elapsed 99%CPU
  0.00user 8.34system 0:08.35elapsed 99%CPU

2)After
  0.00user 1.10system 0:01.10elapsed 99%CPU
  0.00user 0.00system 0:00.01elapsed 64%CPU
  0.00user 0.01system 0:00.01elapsed 82%CPU
  0.00user 0.00system 0:00.01elapsed 64%CPU
  0.00user 0.01system 0:00.01elapsed 82%CPU

The results show the performance increases at least in 548 times.

Shakeel Butt tested this patchset with fork-bomb on his configuration:

 > I created 255 memcgs, 255 ext4 mounts and made each memcg create a
 > file containing few KiBs on corresponding mount. Then in a separate
 > memcg of 200 MiB limit ran a fork-bomb.
 >
 > I ran the "perf record -ag -- sleep 60" and below are the results:
 >
 > Without the patch series:
 > Samples: 4M of event 'cycles', Event count (approx.): 3279403076005
 > +  36.40%            fb.sh  [kernel.kallsyms]    [k] shrink_slab
 > +  18.97%            fb.sh  [kernel.kallsyms]    [k] list_lru_count_one
 > +   6.75%            fb.sh  [kernel.kallsyms]    [k] super_cache_count
 > +   0.49%            fb.sh  [kernel.kallsyms]    [k] down_read_trylock
 > +   0.44%            fb.sh  [kernel.kallsyms]    [k] mem_cgroup_iter
 > +   0.27%            fb.sh  [kernel.kallsyms]    [k] up_read
 > +   0.21%            fb.sh  [kernel.kallsyms]    [k] osq_lock
 > +   0.13%            fb.sh  [kernel.kallsyms]    [k] shmem_unused_huge_count
 > +   0.08%            fb.sh  [kernel.kallsyms]    [k] shrink_node_memcg
 > +   0.08%            fb.sh  [kernel.kallsyms]    [k] shrink_node
 >
 > With the patch series:
 > Samples: 4M of event 'cycles', Event count (approx.): 2756866824946
 > +  47.49%            fb.sh  [kernel.kallsyms]    [k] down_read_trylock
 > +  30.72%            fb.sh  [kernel.kallsyms]    [k] up_read
 > +   9.51%            fb.sh  [kernel.kallsyms]    [k] mem_cgroup_iter
 > +   1.69%            fb.sh  [kernel.kallsyms]    [k] shrink_node_memcg
 > +   1.35%            fb.sh  [kernel.kallsyms]    [k] mem_cgroup_protected
 > +   1.05%            fb.sh  [kernel.kallsyms]    [k] queued_spin_lock_slowpath
 > +   0.85%            fb.sh  [kernel.kallsyms]    [k] _raw_spin_lock
 > +   0.78%            fb.sh  [kernel.kallsyms]    [k] lruvec_lru_size
 > +   0.57%            fb.sh  [kernel.kallsyms]    [k] shrink_node
 > +   0.54%            fb.sh  [kernel.kallsyms]    [k] queue_work_on
 > +   0.46%            fb.sh  [kernel.kallsyms]    [k] shrink_slab_memcg

[ktkhai@virtuozzo.com: v9]
Link: http://lkml.kernel.org/r/153112561772.4097.11011071937553113003.stgit@localhost.localdomain
Link: http://lkml.kernel.org/r/153063070859.1818.11870882950920963480.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Tested-by: Shakeel Butt <shakeelb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matthias Kaehlcke <mka@chromium.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sahitya Tummala <stummala@codeaurora.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: add SHRINK_EMPTY shrinker methods return value

We need to distinguish the situations when shrinker has very small
amount of objects (see vfs_pressure_ratio() called from
super_cache_count()), and when it has no objects at all.  Currently, in
the both of these cases, shrinker::count_objects() returns 0.

The patch introduces new SHRINK_EMPTY return value, which will be used
for "no objects at all" case.  It's is a refactoring mostly, as
SHRINK_EMPTY is replaced by 0 by all callers of do_shrink_slab() in this
patch, and all the magic will happen in further.

Link: http://lkml.kernel.org/r/153063069574.1818.11037751256699341813.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Tested-by: Shakeel Butt <shakeelb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matthias Kaehlcke <mka@chromium.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sahitya Tummala <stummala@codeaurora.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan.c: generalize shrink_slab() calls in shrink_node()

The patch makes shrink_slab() be called for root_mem_cgroup in the same
way as it's called for the rest of cgroups.  This simplifies the logic
and improves the readability.

[ktkhai@virtuozzo.com: wrote changelog]
Link: http://lkml.kernel.org/r/153063068338.1818.11496084754797453962.stgit@localhost.localdomain
Signed-off-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Tested-by: Shakeel Butt <shakeelb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matthias Kaehlcke <mka@chromium.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sahitya Tummala <stummala@codeaurora.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/vmscan.c: iterate only over charged shrinkers during memcg shrink_slab()

Using the preparations made in previous patches, in case of memcg
shrink, we may avoid shrinkers, which are not set in memcg's shrinkers
bitmap.  To do that, we separate iterations over memcg-aware and
!memcg-aware shrinkers, and memcg-aware shrinkers are chosen via
for_each_set_bit() from the bitmap.  In case of big nodes, having many
isolated environments, this gives significant performance growth.  See
next patches for the details.

Note that the patch does not respect to empty memcg shrinkers, since we
never clear the bitmap bits after we set it once.  Their shrinkers will
be called again, with no shrinked objects as result.  This functionality
is provided by next patches.

[ktkhai@virtuozzo.com: v9]
Link: http://lkml.kernel.org/r/153112558507.4097.12713813335683345488.stgit@localhost.localdomain
Link: http://lkml.kernel.org/r/153063066653.1818.976035462801487910.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Tested-by: Shakeel Butt <shakeelb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matthias Kaehlcke <mka@chromium.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sahitya Tummala <stummala@codeaurora.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/list_lru.c: set bit in memcg shrinker bitmap on first list_lru item appearance

Introduce set_shrinker_bit() function to set shrinker-related bit in
memcg shrinker bitmap, and set the bit after the first item is added and
in case of reparenting destroyed memcg's items.

This will allow next patch to make shrinkers be called only, in case of
they have charged objects at the moment, and to improve shrink_slab()
performance.

[ktkhai@virtuozzo.com: v9]
Link: http://lkml.kernel.org/r/153112557572.4097.17315791419810749985.stgit@localhost.localdomain
Link: http://lkml.kernel.org/r/153063065671.1818.15914674956134687268.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Tested-by: Shakeel Butt <shakeelb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matthias Kaehlcke <mka@chromium.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sahitya Tummala <stummala@codeaurora.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memcontrol.c: export mem_cgroup_is_root()

This will be used in next patch.

Link: http://lkml.kernel.org/r/153063064347.1818.1987011484100392706.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Tested-by: Shakeel Butt <shakeelb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matthias Kaehlcke <mka@chromium.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sahitya Tummala <stummala@codeaurora.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/list_lru.c: pass lru argument to memcg_drain_list_lru_node()

This is just refactoring to allow next patches to have lru pointer in
memcg_drain_list_lru_node().

Link: http://lkml.kernel.org/r/153063063164.1818.55009531386089350.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Tested-by: Shakeel Butt <shakeelb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matthias Kaehlcke <mka@chromium.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sahitya Tummala <stummala@codeaurora.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/list_lru: pass dst_memcg argument to memcg_drain_list_lru_node()

This is just refactoring to allow the next patches to have dst_memcg
pointer in memcg_drain_list_lru_node().

Link: http://lkml.kernel.org/r/153063062118.1818.2761273817739499749.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Tested-by: Shakeel Butt <shakeelb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matthias Kaehlcke <mka@chromium.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sahitya Tummala <stummala@codeaurora.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/list_lru.c: add memcg argument to list_lru_from_kmem()

This is just refactoring to allow the next patches to have memcg pointer
in list_lru_from_kmem().

Link: http://lkml.kernel.org/r/153063060664.1818.9541345386733498582.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Tested-by: Shakeel Butt <shakeelb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matthias Kaehlcke <mka@chromium.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sahitya Tummala <stummala@codeaurora.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs: propagate shrinker::id to list_lru

Add list_lru::shrinker_id field and populate it by registered shrinker
id.

This will be used to set correct bit in memcg shrinkers map by lru code
in next patches, after there appeared the first related to memcg element
in list_lru.

Link: http://lkml.kernel.org/r/153063059758.1818.14866596416857717800.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Tested-by: Shakeel Butt <shakeelb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matthias Kaehlcke <mka@chromium.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sahitya Tummala <stummala@codeaurora.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs/super.c: refactor alloc_super()

Do two list_lru_init_memcg() calls after prealloc_super().
destroy_unused_super() in fail path is OK with this.  Next patch needs
such the order.

Link: http://lkml.kernel.org/r/153063058712.1818.3382490999719078571.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Tested-by: Shakeel Butt <shakeelb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matthias Kaehlcke <mka@chromium.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sahitya Tummala <stummala@codeaurora.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/workingset.c: refactor workingset_init()

Use prealloc_shrinker()/register_shrinker_prepared() instead of
register_shrinker().  This will be used in next patch.

[ktkhai@virtuozzo.com: v9]
Link: http://lkml.kernel.org/r/153112550112.4097.16606173020912323761.stgit@localhost.localdomain
Link: http://lkml.kernel.org/r/153063057666.1818.17625951186610808734.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Tested-by: Shakeel Butt <shakeelb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matthias Kaehlcke <mka@chromium.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sahitya Tummala <stummala@codeaurora.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm, memcg: assign memcg-aware shrinkers bitmap to memcg

Imagine a big node with many cpus, memory cgroups and containers.  Let
we have 200 containers, every container has 10 mounts, and 10 cgroups.
All container tasks don't touch foreign containers mounts.  If there is
intensive pages write, and global reclaim happens, a writing task has to
iterate over all memcgs to shrink slab, before it's able to go to
shrink_page_list().

Iteration over all the memcg slabs is very expensive: the task has to
visit 200 * 10 = 2000 shrinkers for every memcg, and since there are
2000 memcgs, the total calls are 2000 * 2000 = 4000000.

So, the shrinker makes 4 million do_shrink_slab() calls just to try to
isolate SWAP_CLUSTER_MAX pages in one of the actively writing memcg via
shrink_page_list().  I've observed a node spending almost 100% in
kernel, making useless iteration over already shrinked slab.

This patch adds bitmap of memcg-aware shrinkers to memcg.  The size of
the bitmap depends on bitmap_nr_ids, and during memcg life it's
maintained to be enough to fit bitmap_nr_ids shrinkers.  Every bit in
the map is related to corresponding shrinker id.

Next patches will maintain set bit only for really charged memcg.  This
will allow shrink_slab() to increase its performance in significant way.
See the last patch for the numbers.

[ktkhai@virtuozzo.com: v9]
Link: http://lkml.kernel.org/r/153112549031.4097.3576147070498769979.stgit@localhost.localdomain
[ktkhai@virtuozzo.com: add comment to mem_cgroup_css_online()]
Link: http://lkml.kernel.org/r/521f9e5f-c436-b388-fe83-4dc870bfb489@virtuozzo.com
Link: http://lkml.kernel.org/r/153063056619.1818.12550500883688681076.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Tested-by: Shakeel Butt <shakeelb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matthias Kaehlcke <mka@chromium.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sahitya Tummala <stummala@codeaurora.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/memcontrol.c: move up for_each_mem_cgroup{, _tree} defines

Next patch requires these defines are above their current position, so
here they are moved to declarations.

Link: http://lkml.kernel.org/r/153063055665.1818.5200425793649695598.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Tested-by: Shakeel Butt <shakeelb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matthias Kaehlcke <mka@chromium.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sahitya Tummala <stummala@codeaurora.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: assign id to every memcg-aware shrinker

Introduce shrinker::id number, which is used to enumerate memcg-aware
shrinkers.  The number start from 0, and the code tries to maintain it
as small as possible.

This will be used to represent a memcg-aware shrinkers in memcg
shrinkers map.

Since all memcg-aware shrinkers are based on list_lru, which is
per-memcg in case of !CONFIG_MEMCG_KMEM only, the new functionality will
be under this config option.

[ktkhai@virtuozzo.com: v9]
Link: http://lkml.kernel.org/r/153112546435.4097.10607140323811756557.stgit@localhost.localdomain
Link: http://lkml.kernel.org/r/153063054586.1818.6041047871606697364.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Tested-by: Shakeel Butt <shakeelb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matthias Kaehlcke <mka@chromium.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sahitya Tummala <stummala@codeaurora.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm: introduce CONFIG_MEMCG_KMEM as combination of CONFIG_MEMCG && !CONFIG_SLOB

Introduce new config option, which is used to replace repeating
CONFIG_MEMCG && !CONFIG_SLOB pattern.  Next patches add a little more
memcg+kmem related code, so let's keep the defines more clearly.

Link: http://lkml.kernel.org/r/153063053670.1818.15013136946600481138.stgit@localhost.localdomain
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Tested-by: Shakeel Butt <shakeelb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Josef Bacik <jbacik@fb.com>
Cc: Li RongQing <lirongqing@baidu.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Matthias Kaehlcke <mka@chromium.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Philippe Ombredanne <pombredanne@nexb.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sahitya Tummala <stummala@codeaurora.org>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>