find_idlest_group() only compares the runnable_load_avg when looking
for the least loaded group. But on fork intensive use case like
hackbench where tasks blocked quickly after the fork, this can lead to
selecting the same CPU instead of other CPUs, which have similar
runnable load but a lower load_avg.
When the runnable_load_avg of 2 CPUs are close, we now take into
account the amount of blocked load as a 2nd selection factor. There is
now 3 zones for the runnable_load of the rq:
- [0 .. (runnable_load - imbalance)]:
Select the new rq which has significantly less runnable_load
- [(runnable_load - imbalance) .. (runnable_load + imbalance)]:
The runnable loads are close so we use load_avg to chose
between the 2 rq
- [(runnable_load + imbalance) .. ULONG_MAX]:
Keep the current rq which has significantly less runnable_load
The scale factor that is currently used for comparing runnable_load,
doesn't work well with small value. As an example, the use of a
scaling factor fails as soon as this_runnable_load == 0 because we
always select local rq even if min_runnable_load is only 1, which
doesn't really make sense because they are just the same. So instead
of scaling factor, we use an absolute margin for runnable_load to
detect CPUs with similar runnable_load and we keep using scaling
factor for blocked load.
For use case like hackbench, this enable the scheduler to select
different CPUs during the fork sequence and to spread tasks across the
system.
Tests have been done on a Hikey board (ARM based octo cores) for
several kernel. The result below gives min, max, avg and stdev values
of 18 runs with each configuration.
The patches depend on the "no missing update_rq_clock()" work.
hackbench -P -g 1
ea86cb4b767dc603c902 v4.8 v4.8+patches
min 0.049 0.050 0.051 0,048
avg 0.057 0.057(0%) 0.057(0%) 0,055(+5%)
max 0.066 0.068 0.070 0,063
stdev +/-9% +/-9% +/-8% +/-9%
More performance numbers here:
https://lkml.kernel.org/r/20161203214707.GI20785@codeblueprint.co.uk
Tested-by: Matt Fleming <matt@codeblueprint.co.uk>
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Matt Fleming <matt@codeblueprint.co.uk>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Morten.Rasmussen@arm.com
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: dietmar.eggemann@arm.com
Cc: kernellwp@gmail.com
Cc: umgwanakikbuti@gmail.com
Cc: yuyang.du@intel.comc
Link: http://lkml.kernel.org/r/1481216215-24651-3-git-send-email-vincent.guittot@linaro.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
During fork, the utilization of a task is init once the rq has been
selected because the current utilization level of the rq is used to
set the utilization of the fork task. As the task's utilization is
still 0 at this step of the fork sequence, it doesn't make sense to
look for some spare capacity that can fit the task's utilization.
Furthermore, I can see perf regressions for the test:
hackbench -P -g 1
because the least loaded policy is always bypassed and tasks are not
spread during fork.
With this patch and the fix below, we are back to same performances as
for v4.8. The fix below is only a temporary one used for the test
until a smarter solution is found because we can't simply remove the
test which is useful for others benchmarks
| @@ -5708,13 +5708,6 @@ static int select_idle_cpu(struct task_struct *p, struct sched_domain *sd, int t
|
| avg_cost = this_sd->avg_scan_cost;
|
| - /*
| - * Due to large variance we need a large fuzz factor; hackbench in
| - * particularly is sensitive here.
| - */
| - if ((avg_idle / 512) < avg_cost)
| - return -1;
| -
| time = local_clock();
|
| for_each_cpu_wrap(cpu, sched_domain_span(sd), target, wrap) {
Tested-by: Matt Fleming <matt@codeblueprint.co.uk>
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Matt Fleming <matt@codeblueprint.co.uk>
Acked-by: Morten Rasmussen <morten.rasmussen@arm.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: dietmar.eggemann@arm.com
Cc: kernellwp@gmail.com
Cc: umgwanakikbuti@gmail.com
Cc: yuyang.du@intel.comc
Link: http://lkml.kernel.org/r/1481216215-24651-2-git-send-email-vincent.guittot@linaro.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Commit:
3cded41794 ("x86/paravirt: Optimize native pv_lock_ops.vcpu_is_preempted()")
introduced a paravirt op with bool return type [*]
It turns out that the PVOP_CALL*() macros miscompile when rettype is
bool. Code that looked like:
83 ef 01 sub $0x1,%edi
ff 15 32 a0 d8 00 callq *0xd8a032(%rip) # ffffffff81e28120 <pv_lock_ops+0x20>
84 c0 test %al,%al
ended up looking like so after PVOP_CALL1() was applied:
83 ef 01 sub $0x1,%edi
48 63 ff movslq %edi,%rdi
ff 14 25 20 81 e2 81 callq *0xffffffff81e28120
48 85 c0 test %rax,%rax
Note how it tests the whole of %rax, even though a typical bool return
function only sets %al, like:
0f 95 c0 setne %al
c3 retq
This is because ____PVOP_CALL() does:
__ret = (rettype)__eax;
and while regular integer type casts truncate the result, a cast to
bool tests for any !0 value. Fix this by explicitly truncating to
sizeof(rettype) before casting.
[*] The actual bug should've been exposed in commit:
446f3dc8cc ("locking/core, x86/paravirt: Implement vcpu_is_preempted(cpu) for KVM and Xen guests")
but that didn't properly implement the paravirt call.
Reported-by: kernel test robot <xiaolong.ye@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alok Kataria <akataria@vmware.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Chris Wright <chrisw@sous-sol.org>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Pan Xinhui <xinhui.pan@linux.vnet.ibm.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Peter Anvin <hpa@zytor.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Thomas Gleixner <tglx@linutronix.de>
Fixes: 3cded41794 ("x86/paravirt: Optimize native pv_lock_ops.vcpu_is_preempted()")
Link: http://lkml.kernel.org/r/20161208154349.346057680@infradead.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
While chasing a regression I noticed we potentially patch the wrong
code in native_patch().
If we do not select the native code sequence, we must use the default
patcher, not fall-through the switch case.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alok Kataria <akataria@vmware.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Chris Wright <chrisw@sous-sol.org>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Pan Xinhui <xinhui.pan@linux.vnet.ibm.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Peter Anvin <hpa@zytor.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: kernel test robot <xiaolong.ye@intel.com>
Fixes: 3cded41794 ("x86/paravirt: Optimize native pv_lock_ops.vcpu_is_preempted()")
Link: http://lkml.kernel.org/r/20161208154349.270616999@infradead.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
An earlier patch allowed enabling PT and LBR at the same
time on Goldmont. However it also allowed enabling BTS and LBR
at the same time, which is still not supported. Fix this by
bypassing the check only for PT.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: alexander.shishkin@intel.com
Cc: kan.liang@intel.com
Cc: <stable@vger.kernel.org>
Fixes: ccbebba4c6 ("perf/x86/intel/pt: Bypass PT vs. LBR exclusivity if the core supports it")
Link: http://lkml.kernel.org/r/20161209001417.4713-1-andi@firstfloor.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The hardware documentation says bit 11:10 are used for the GPE
frequency selection. Fix the mask in the define to match these bits.
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Thomas Langer <thomas.langer@intel.com>
Cc: linux-mips@linux-mips.org
Cc: john@phrozen.org
Patchwork: https://patchwork.linux-mips.org/patch/14648/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The sync_cmos_clock function in kernel/time/ntp.c first tries to update
the internal clock of the cpu by calling the "update_persistent_clock64"
architecture specific function. If this returns -ENODEV, it then tries
to update an external RTC using "rtc_set_ntp_time".
On the mips architecture, the weak implementation of the underlying
function would return 0 if it wasn't overridden. This meant that the
sync_cmos_clock function would never try to update an external RTC
(if both CONFIG_GENERIC_CMOS_UPDATE and CONFIG_RTC_SYSTOHC are
configured)
Returning -ENODEV instead, means that an external RTC will be tried.
Signed-off-by: Luuk Paulussen <luuk.paulussen@alliedtelesis.co.nz>
Reviewed-by: Richard Laing <richard.laing@alliedtelesis.co.nz>
Reviewed-by: Scott Parlane <scott.parlane@alliedtelesis.co.nz>
Reviewed-by: Chris Packham <chris.packham@alliedtelesis.co.nz>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/14649/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
In commit 02cea39586 ("genirq: Provide disable_hardirq()")
Peter introduced disable_hardirq() for netpoll, but it is forgotten
to use it for e1000.
This patch changes disable_irq() to disable_hardirq() for e1000.
Reported-by: Dave Jones <davej@codemonkey.org.uk>
Suggested-by: Sabrina Dubroca <sd@queasysnail.net>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The .match_method field is a u8, so we shouldn't be casting to a u16,
and because it is only one byte, we do not need to byte swap anything.
Just assign the value directly. This avoids issues on Big Endian
architectures which would have byte swapped and then incorrectly
truncated the value.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Bimmy Pujari <bimmy.pujari@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds support of the cpts device found in the
gbe and 10gbe ethernet switches on the keystone 2 SoCs
(66AK2E/L/Hx, 66AK2Gx).
Cc: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: WingMan Kwok <w-kwok2@ti.com>
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of having individual PHY drivers set the SUPPORTED_Pause and
SUPPORTED_Asym_Pause flags, phylib itself should set those flags,
unless there is a hardware erratum or other special case. During
autonegotiation, the PHYs will determine whether to enable pause
frame support.
Pause frames are a feature that is supported by the MAC. It is the MAC
that generates the frames and that processes them. The PHY can only be
configured to allow them to pass through.
This commit also effectively reverts the recently applied c7a61319
("net: phy: dp83848: Support ethernet pause frames").
So the new process is:
1) Unless the PHY driver overrides it, phylib sets the SUPPORTED_Pause
and SUPPORTED_AsymPause bits in phydev->supported. This indicates that
the PHY supports pause frames.
2) The MAC driver checks phydev->supported before it calls phy_start().
If (SUPPORTED_Pause | SUPPORTED_AsymPause) is set, then the MAC driver
sets those bits in phydev->advertising, if it wants to enable pause
frame support.
3) When the link state changes, the MAC driver checks phydev->pause and
phydev->asym_pause, If the bits are set, then it enables the corresponding
features in the MAC. The algorithm is:
if (phydev->pause)
The MAC should be programmed to receive and honor
pause frames it receives, i.e. enable receive flow control.
if (phydev->pause != phydev->asym_pause)
The MAC should be programmed to transmit pause
frames when needed, i.e. enable transmit flow control.
Signed-off-by: Timur Tabi <timur@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
PPPOL2TP_MSG_* and L2TP_MSG_* are duplicates, and are being used
interchangeably in the kernel, so let's standardize on L2TP_MSG_*
internally, and keep PPPOL2TP_MSG_* defined in UAPI for compatibility.
Signed-off-by: Asbjoern Sloth Toennesen <asbjorn@asbjorn.st>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move the L2TP_MSG_* definitions to UAPI, as it is part of
the netlink API.
Signed-off-by: Asbjoern Sloth Toennesen <asbjorn@asbjorn.st>
Signed-off-by: David S. Miller <davem@davemloft.net>
Lino Sanfilippo says:
====================
Remove private tx queue locks
this patch series removes unnecessary private locks in the sxgbe and the
stmmac driver.
v2:
- adjust commit message
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The driver uses a private lock for synchronization of the xmit function and
the xmit completion handler, but since the NETIF_F_LLTX flag is not set,
the xmit function is also called with the xmit_lock held.
On the other hand the completion handler uses the reverse locking order by
first taking the private lock and (in case that the tx queue had been
stopped) then the xmit_lock.
Improve the locking by removing the private lock and using only the
xmit_lock for synchronization instead.
Signed-off-by: Lino Sanfilippo <LinoSanfilippo@gmx.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The driver uses a private lock for synchronization of the xmit function and
the xmit completion handler, but since the NETIF_F_LLTX flag is not set,
the xmit function is also called with the xmit_lock held.
On the other hand the completion handler uses the reverse locking order by
first taking the private lock and (in case that the tx queue had been
stopped) then the xmit_lock.
Improve the locking by removing the private lock and using only the
xmit_lock for synchronization instead.
Signed-off-by: Lino Sanfilippo <LinoSanfilippo@gmx.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Vivien Didelot says:
====================
net: bridge: fast ageing on topology change
802.1D [1] specifies that the bridges in a network must use a short
value to age out dynamic entries in the Filtering Database for a period,
once a topology change has been communicated by the root bridge.
This patchset fixes this for the in-kernel STP implementation.
Once the topology change flag is set in a net_bridge instance, the
ageing time value is shorten to twice the forward delay used by the
topology.
When the topology change flag is cleared, the ageing time configured for
the bridge is restored.
To accomplish that, a new bridge_ageing_time member is added to the
net_bridge structure, to store the user configured bridge ageing time.
Two helpers are added to offload the ageing time and set the topology
change flag in the net_bridge instance. Then the required logic is added
in the topology change helper if in-kernel STP is used.
This has been tested on the following topology:
+--------------+
| root bridge |
| 1 2 3 4 |
+--+--+--+--+--+
| | | | +--------+
| | | +------| laptop |
| | | +--------+
+--+--+--+-----+
| 1 2 3 |
| slave bridge |
+--------------+
When unplugging/replugging the laptop, the slave bridge (under test)
gets the topology change flag sent by the root bridge, and fast ageing
is triggered on the bridges. Once the topology change timer of the root
bridge expires, the topology change flag is cleared and the configured
ageing time is restored on the bridges.
A similar test has been done between two bridges under test.
When changing the forward delay of the root bridge with:
# echo 3000 > /sys/class/net/br0/bridge/forward_delay
the ageing time correctly changes on both bridges from 300s to 60s while
the TOPOLOGY_CHANGE flag is present.
[1] "8.3.5 Notifying topology changes",
http://profesores.elo.utfsm.cl/~agv/elo309/doc/802.1D-1998.pdf
No change since RFC: https://lkml.org/lkml/2016/10/19/828
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
802.1D [1] specifies that the bridges must use a short value to age out
dynamic entries in the Filtering Database for a period, once a topology
change has been communicated by the root bridge.
Add a bridge_ageing_time member in the net_bridge structure to store the
bridge ageing time value configured by the user (ioctl/netlink/sysfs).
If we are using in-kernel STP, shorten the ageing time value to twice
the forward delay used by the topology when the topology change flag is
set. When the flag is cleared, restore the configured ageing time.
[1] "8.3.5 Notifying topology changes ",
http://profesores.elo.utfsm.cl/~agv/elo309/doc/802.1D-1998.pdf
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a __br_set_topology_change helper to set the topology change value.
This can be later extended to add actions when the topology change flag
is set or cleared.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The SWITCHDEV_ATTR_ID_BRIDGE_AGEING_TIME switchdev attr is actually set
when initializing a bridge port, and when configuring the bridge ageing
time from ioctl/netlink/sysfs.
Add a __set_ageing_time helper to offload the ageing time to physical
switches, and add the SWITCHDEV_F_DEFER flag since it can be called
under bridge lock.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ethtool api {get|set}_settings is deprecated.
We move this driver to new api {get|set}_link_ksettings.
Signed-off-by: Philippe Reynes <tremyfr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The channels are common for both ndevs in dual emac mode. Hence, keep
in sync their rates.
Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Don't re-split res in the following cases:
- speed of phys is not changed
- speed of phys is changed and no rate limited channels
- speed of phys is changed and all channels are rate limited
- phy is unlinked while dev is open
- phy is linked back but speed is not changed
The maximum speed is sum of "linked" phys, thus res are split taken
in account two interfaces, both for dual emac mode and for
switch mode.
Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Re-split weight along with budget. It simplify code a little
and update state after every rate change. Also it's necessarily
to move arguments checks to this combined function. Replace
maximum rate check for an interface on maximum possible rate.
Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
No need to start queues after cpsw is started as it will be done
while cpsw_adjust_link(), after phy connection.
Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use the same, more convenient macros, to get active slave.
Signed-off-by: Ivan Khoronzhuk <ivan.khoronzhuk@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
We previously relied on GENERIC_ALLOCATOR to be selected by CONFIG_ARM,
but now we can compile-test the driver on other architectures that
don't select it:
drivers/net/built-in.o: In function `mvneta_bm_remove':
mvneta_bm.c:(.text+0x4ee35): undefined reference to `gen_pool_free'
This adds an explicit select for the part of the driver that has
the dependency.
Fixes: a0627f776a ("net: marvell: Allow drivers to be built with COMPILE_TEST")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch removes a newline which was added
in socket.c file in net-next
Signed-off-by: Amit Kushwaha <kushwaha.a@samsung.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
netlink_chain is called in ->release(), which is apparently
a process context, so we don't have to use an atomic notifier
here.
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull crypto fixes from Herbert Xu:
"This fixes the following issues:
- Fix pointer size when caam is used with AArch64 boot loader on
AArch32 kernel.
- Fix ahash state corruption in marvell driver.
- Fix buggy algif_aed tag handling.
- Prevent mcryptd from being used with incompatible algorithms which
can cause crashes"
* 'linus' of git://git.kernel.org/pub/scm/linux/kernel/git/herbert/crypto-2.6:
crypto: algif_aead - fix uninitialized variable warning
crypto: mcryptd - Check mcryptd algorithm compatibility
crypto: algif_aead - fix AEAD tag memory handling
crypto: caam - fix pointer size for AArch64 boot loader, AArch32 kernel
crypto: marvell - Don't corrupt state of an STD req for re-stepped ahash
crypto: marvell - Don't copy hash operation twice into the SRAM
Pull networking fixes from David Miller:
1) Limit the number of can filters to avoid > MAX_ORDER allocations.
Fix from Marc Kleine-Budde.
2) Limit GSO max size in netvsc driver to avoid problems with NVGRE
configurations. From Stephen Hemminger.
3) Return proper error when memory allocation fails in
ser_gigaset_init(), from Dan Carpenter.
4) Missing linkage undo in error paths of ipvlan_link_new(), from Gao
Feng.
5) Missing necessayr SET_NETDEV_DEV in lantiq and cpmac drivers, from
Florian Fainelli.
6) Handle probe deferral properly in smsc911x driver.
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net:
net: mlx5: Fix Kconfig help text
net: smsc911x: back out silently on probe deferrals
ibmveth: set correct gso_size and gso_type
net: ethernet: cpmac: Call SET_NETDEV_DEV()
net: ethernet: lantiq_etop: Call SET_NETDEV_DEV()
vhost-vsock: fix orphan connection reset
cxgb4/cxgb4vf: Assign netdev->dev_port with port ID
driver: ipvlan: Unlink the upper dev when ipvlan_link_new failed
ser_gigaset: return -ENOMEM on error instead of success
NET: usb: cdc_mbim: add quirk for supporting Telit LE922A
can: peak: fix bad memory access and free sequence
phy: Don't increment MDIO bus refcount unless it's a different owner
netvsc: reduce maximum GSO size
drivers: net: cpsw-phy-sel: Clear RGMII_IDMODE on "rgmii" links
can: raw: raw_setsockopt: limit number of can_filter that can be set
Since the following commit, Infiniband and Ethernet have not been
mutually exclusive.
Fixes: 4aa17b28 mlx5: Enable mutual support for IB and Ethernet
Signed-off-by: Christopher Covington <cov@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
It seems attackers can also send UDP packets with no payload at all.
skb_condense() can still be a win in this case.
It will be possible to replace the custom code in tcp_add_backlog()
to get full benefit from skb_condense()
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When trying to get a regulator we may get deferred and we see
this noise:
smsc911x 1b800000.ethernet-ebi2 (unnamed net_device) (uninitialized):
couldn't get regulators -517
Then the driver continues anyway. Which means that the regulator
may not be properly retrieved and reference counted, and may be
switched off in case noone else is using it.
Fix this by returning silently on deferred probe and let the
system work it out.
Cc: Jeremy Linton <jeremy.linton@arm.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
* fix a logic bug introduced by a previous cleanup
* fix nl80211 attribute confusing (trying to use
a single attribute for two purposes)
* fix a long-standing BSS leak that happens when an
association attempt is abandoned
-----BEGIN PGP SIGNATURE-----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=HvKt
-----END PGP SIGNATURE-----
Merge tag 'mac80211-next-for-davem-2016-12-09' of git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211-next
Johannes Berg says:
====================
Three fixes:
* fix a logic bug introduced by a previous cleanup
* fix nl80211 attribute confusing (trying to use
a single attribute for two purposes)
* fix a long-standing BSS leak that happens when an
association attempt is abandoned
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch is based on an earlier one submitted
by Jon Maxwell with the following commit message:
"We recently encountered a bug where a few customers using ibmveth on the
same LPAR hit an issue where a TCP session hung when large receive was
enabled. Closer analysis revealed that the session was stuck because the
one side was advertising a zero window repeatedly.
We narrowed this down to the fact the ibmveth driver did not set gso_size
which is translated by TCP into the MSS later up the stack. The MSS is
used to calculate the TCP window size and as that was abnormally large,
it was calculating a zero window, even although the sockets receive buffer
was completely empty."
We rely on the Virtual I/O Server partition in a pseries
environment to provide the MSS through the TCP header checksum
field. The stipulation is that users should not disable checksum
offloading if rx packet aggregation is enabled through VIOS.
Some firmware offerings provide the MSS in the RX buffer.
This is signalled by a bit in the RX queue descriptor.
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Reviewed-by: Pradeep Satyanarayana <pradeeps@linux.vnet.ibm.com>
Reviewed-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Reviewed-by: Jonathan Maxwell <jmaxwell37@gmail.com>
Reviewed-by: David Dai <zdai@us.ibm.com>
Signed-off-by: Thomas Falcon <tlfalcon@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Eric Dumazet says:
====================
udp: receive path optimizations
This patch series provides about 100 % performance increase under flood.
v2: added Paolo feedback on udp_rmem_release() for tiny sk_rcvbuf
added the last patch touching sk_rmem_alloc later
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
In flood situations, keeping sk_rmem_alloc at a high value
prevents producers from touching the socket.
It makes sense to lower sk_rmem_alloc only at the end
of udp_rmem_release() after the thread draining receive
queue in udp_recvmsg() finished the writes to sk_forward_alloc.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If udp_recvmsg() constantly releases sk_rmem_alloc
for every read packet, it gives opportunity for
producers to immediately grab spinlocks and desperatly
try adding another packet, causing false sharing.
We can add a simple heuristic to give the signal
by batches of ~25 % of the queue capacity.
This patch considerably increases performance under
flood by about 50 %, since the thread draining the queue
is no longer slowed by false sharing.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In UDP RX handler, we currently clear skb->dev before skb
is added to receive queue, because device pointer is no longer
available once we exit from RCU section.
Since this first cache line is always hot, lets reuse this space
to store skb->truesize and thus avoid a cache line miss at
udp_recvmsg()/udp_skb_destructor time while receive queue
spinlock is held.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Idea of busylocks is to let producers grab an extra spinlock
to relieve pressure on the receive_queue spinlock shared by consumer.
This behavior is requested only once socket receive queue is above
half occupancy.
Under flood, this means that only one producer can be in line
trying to acquire the receive_queue spinlock.
These busylock can be allocated on a per cpu manner, instead of a
per socket one (that would consume a cache line per socket)
This patch considerably improves UDP behavior under stress,
depending on number of NIC RX queues and/or RPS spread.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Timur Tabi says:
====================
net: qcom/emac: simplify support for different SOCs
On SOCs that have the Qualcomm EMAC network controller, the internal
PHY block is always different. Sometimes the differences are small,
sometimes it might be a completely different IP. Either way, using version
numbers to differentiate them and putting all of the init code in one
file does not scale.
This patchset does two things: The first breaks up the current code into
different files, and the second patch adds support for a third SOC, the
Qualcomm Technologies QDF2400 ARM Server SOC.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
The QDF2432 and the QDF2400 have slightly different internal PHYs,
so there are some programming differences. Some of the registers in
the QDF2400 have moved, and some registers require different values
during initialization.
Because of the differences, and because HIDs are a scare resource,
the ACPI tables specify the hardware version in an _HRV property.
Version 1 is the QDF2432, and version 2 is the QDF2400. Any future
SOC that has the same internal PHY but different programming
requirements will be assigned the next available version number.
Signed-off-by: Timur Tabi <timur@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The internal PHY of the EMAC differs on each SOC, and the list will
only continue to grow. By separating the code into individual files,
we can add support for more SOCs more cleanly.
Note: The internal PHY is also sometimes called the SGMII device.
We also stop referring to the various PHY variations by version number,
so no more "v2", "v3", etc. Instead, the devices are named after the
SOC they are, which is in sync with the device tree property names.
Future patches will probably rearrange more code among the files.
Signed-off-by: Timur Tabi <timur@codeaurora.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
ldt->size can never be negative. The helper functions take 'unsigned int'
arguments which are assigned from ldt->size. The related user space
user_desc struct member entry_number is unsigned as well.
But ldt->size itself and a few local variables which are related to
ldt->size are type 'int' which makes no sense whatsoever and results in
typecasts which make the eyes bleed.
Clean it up and convert everything which is related to ldt->size to
unsigned it.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Dan Carpenter <dan.carpenter@oracle.com>