- 05 Jan, 2010 1 commit
-
-
Catalin(ux) M. BOIE authored
I was very frustrated about the fact that I have to recompile the kernel to change the hash size. So, I created this patch. If IPVS is built-in you can append ip_vs.conn_tab_bits=?? to kernel command line, or, if you built IPVS as modules, you can add options ip_vs conn_tab_bits=??. To keep everything backward compatible, you still can select the size at compile time, and that will be used as default. It has been about a year since this patch was originally posted and subsequently dropped on the basis of insufficient test data. Mark Bergsma has provided the following test results which seem to strongly support the need for larger hash table sizes: We do however run into the same problem with the default setting (212 = 4096 entries), as most of our LVS balancers handle around a million connections/SLAB entries at any point in time (around 100-150 kpps load). With only 4096 hash table entries this implies that each entry consists of a linked list of 256 connections *on average*. To provide some statistics, I did an oprofile run on an 2.6.31 kernel, with both the default 4096 table size, and the same kernel recompiled with IP_VS_CONN_TAB_BITS set to 18 (218 = 262144 entries). I built a quick test setup with a part of Wikimedia/Wikipedia's live traffic mirrored by the switch to the test host. With the default setting, at ~ 120 kpps packet load we saw a typical %si CPU usage of around 30-35%, and oprofile reported a hot spot in ip_vs_conn_in_get: samples % image name app name symbol name 1719761 42.3741 ip_vs.ko ip_vs.ko ip_vs_conn_in_get 302577 7.4554 bnx2 bnx2 /bnx2 181984 4.4840 vmlinux vmlinux __ticket_spin_lock 128636 3.1695 vmlinux vmlinux ip_route_input 74345 1.8318 ip_vs.ko ip_vs.ko ip_vs_conn_out_get 68482 1.6874 vmlinux vmlinux mwait_idle After loading the recompiled kernel with 218 entries, %si CPU usage dropped in half to around 12-18%, and oprofile looks much healthier, with only 7% spent in ip_vs_conn_in_get: samples % image name app name symbol name 265641 14.4616 bnx2 bnx2 /bnx2 143251 7.7986 vmlinux vmlinux __ticket_spin_lock 140661 7.6576 ip_vs.ko ip_vs.ko ip_vs_conn_in_get 94364 5.1372 vmlinux vmlinux mwait_idle 86267 4.6964 vmlinux vmlinux ip_route_input [ horms@verge.net.au: trivial up-port and minor style fixes ] Signed-off-by:
Catalin(ux) M. BOIE <catab@embedromix.ro> Cc: Mark Bergsma <mark@wikimedia.org> Signed-off-by:
Simon Horman <horms@verge.net.au> Signed-off-by:
Patrick McHardy <kaber@trash.net>
-
- 04 Jan, 2010 10 commits
-
-
Jan Engelhardt authored
We can initialize the random hash bytes on checkentry. This is preferable since it is outside the hot path. Reference: http://bugzilla.netfilter.org/show_bug.cgi?id=621Signed-off-by:
Jan Engelhardt <jengelh@medozas.de> Signed-off-by:
-
Jan Engelhardt authored
"It is deliberately not done in the init function, since we might not have sufficient random while booting." Signed-off-by:
-
Jan Engelhardt authored
Moving rnd_inited into the hole after the uint8 lets go of the uint32 rnd_inited was using, plus the padding that would follow the int group. Signed-off-by:
-
Julia Lawall authored
obj has type struct snmp_object **, not struct snmp_object *. But indeed it is not even clear why kmalloc is needed. The memory is freed by the end of the function, so the local variable of pointer type should be sufficient. The semantic patch that makes this change is as follows: (http://coccinelle.lip6.fr/) // <smpl> @disable sizeof_type_expr@ type T; T **x; @@ x = <+...sizeof( - T + *x )...+> // </smpl> Signed-off-by:
Julia Lawall <julia@diku.dk> Signed-off-by:
-
Ken Kawasaki authored
axnet_cs: remove unnecessary spin_unlock_irqrestore,spin_lock_irqsave. Signed-off-by:Ken Kawasaki <ken_kawasaki@spring.nifty.jp> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Amerigo Wang authored
We can rely on kconfig to limit these numbers, no need to limit them at compile time/run time. Users who modify these numbers manually should be responsible for themself. :) Signed-off-by:
WANG Cong <amwang@redhat.com> Cc: Per Liden <per.liden@ericsson.com> Cc: Jon Maloy <jon.maloy@ericsson.com> Cc: Allan Stephens <allan.stephens@windriver.com> Cc: David S. Miller <davem@davemloft.net> Signed-off-by:
-
Marc Kleine-Budde authored
This patch adds the flag CAN_CTRLMODE_ONE_SHOT. It is used as mask or flag in the "struct can_ctrlmode". It allows userspace via netlink to set a CAN controller into the special "one-shot" mode. In this mode, if supported by the CAN controller, it tries only once to deliver a CAN frame and aborts it if an error (e.g.: arbitration lost) happens. Signed-off-by:
Marc Kleine-Budde <mkl@pengutronix.de> Acked-by:
Wolfgang Grandegger <wg@grandegger.com> Signed-off-by:
-
Oliver Hartkopp authored
this patch removes the hlist that contains the CAN receiver filter lists. It uses the 'midlayer private' pointer ml_priv and links the filters directly to the CAN netdevice, which allows to omit the walk through the complete CAN devices hlist for each received CAN frame. This patch is tested and does not remove any locking. Signed-off-by:
Oliver Hartkopp <oliver@hartkopp.net> Signed-off-by:
-
Julia Lawall authored
Use kzalloc rather than kcalloc(1,...) The semantic patch that makes this change is as follows: (http://coccinelle.lip6.fr/) // <smpl> @@ @@ - kcalloc(1, + kzalloc( ...) // </smpl> Signed-off-by:
Divy Le Ray <divy@chelsio.com> Signed-off-by:
-
Andy Gospodarek authored
This allows a bond device to specify an arp_ip_target as a host that is not on the same vlan as the base bond device and still use arp validation. A configuration like this, now works: BONDING_OPTS="mode=active-backup arp_interval=1000 arp_ip_target=10.0.100.1 arp_validate=3" 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 16436 qdisc noqueue link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 inet 127.0.0.1/8 scope host lo inet6 ::1/128 scope host valid_lft forever preferred_lft forever 2: eth1: <BROADCAST,MULTICAST,SLAVE,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast master bond0 qlen 1000 link/ether 00:13:21:be:33:e9 brd ff:ff:ff:ff:ff:ff 3: eth0: <BROADCAST,MULTICAST,SLAVE,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast master bond0 qlen 1000 link/ether 00:13:21:be:33:e9 brd ff:ff:ff:ff:ff:ff 8: bond0: <BROADCAST,MULTICAST,MASTER,UP,LOWER_UP> mtu 1500 qdisc noqueue link/ether 00:13:21:be:33:e9 brd ff:ff:ff:ff:ff:ff inet6 fe80::213:21ff:febe:33e9/64 scope link valid_lft forever preferred_lft forever 9: bond0.100@bond0: <BROADCAST,MULTICAST,MASTER,UP,LOWER_UP> mtu 1500 qdisc noqueue link/ether 00:13:21:be:33:e9 brd ff:ff:ff:ff:ff:ff inet 10.0.100.2/24 brd 10.0.100.255 scope global bond0.100 inet6 fe80::213:21ff:febe:33e9/64 scope link valid_lft forever preferred_lft forever Ethernet Channel Bonding Driver: v3.6.0 (September 26, 2009) Bonding Mode: fault-tolerance (active-backup) Primary Slave: None Currently Active Slave: eth1 MII Status: up MII Polling Interval (ms): 0 Up Delay (ms): 0 Down Delay (ms): 0 ARP Polling Interval (ms): 1000 ARP IP target/s (n.n.n.n form): 10.0.100.1 Slave Interface: eth1 MII Status: up Link Failure Count: 1 Permanent HW addr: 00:40:05:30:ff:30 Slave Interface: eth0 MII Status: up Link Failure Count: 0 Permanent HW addr: 00:13:21:be:33:e9 Signed-off-by:Andy Gospodarek <andy@greyhouse.net> Signed-off-by:
Jay Vosburgh <fubar@us.ibm.com> Signed-off-by:
-
- 30 Dec, 2009 14 commits
-
-
-
git://git.kernel.org/pub/scm/linux/kernel/git/linville/wireless-2.6John W. Linville authored
Conflicts: drivers/net/wireless/libertas/scan.c
-
H Hartley Sweeten authored
Use the %pM kernel extension to display the MAC address. Signed-off-by:
H Hartley Sweeten <hsweeten@visionengravers.com> Signed-off-by:
-
H Hartley Sweeten authored
Use the %pM kernel extension to display the MAC address. Signed-off-by:
-
H Hartley Sweeten authored
Use the %pM kernel extension to display the MAC address. Signed-off-by:
-
H Hartley Sweeten authored
Use the %pM kernel extension to display the MAC address. Signed-off-by:
-
H Hartley Sweeten authored
Use the %pM kernel extension to display the MAC address. Signed-off-by:
-
H Hartley Sweeten authored
Use the %pM kernel extension to display the MAC address. Signed-off-by:
-
H Hartley Sweeten authored
Use the %pM kernel extension to display the MAC address. Signed-off-by:
-
H Hartley Sweeten authored
Use the %pM kernel extension to display the MAC address. Signed-off-by:
-
H Hartley Sweeten authored
Use the %pM kernel extension to display the MAC address. Signed-off-by:
-
H Hartley Sweeten authored
Use the %pM kernel extension to display the MAC address. Signed-off-by:
-
H Hartley Sweeten authored
Use the %pM kernel extension to display the MAC address. Signed-off-by:
-
H Hartley Sweeten authored
Use the %pM kernel extension to display the MAC address. The only difference in the output is that the MAC address is shown in the usual colon-separated hex notation instead of space-separated. Signed-off-by:
-
- 29 Dec, 2009 1 commit
-
-
John W. Linville authored
This reverts commit 9bd568a5. That commit is shown to cause allocation failures during initialization on some machines. http://bugzilla.kernel.org/show_bug.cgi?id=14844Signed-off-by:
John W. Linville <linville@tuxdriver.com>
-
- 28 Dec, 2009 14 commits
-
-
Kalle Valo authored
To make it easier to notice cases of calling sleeping ops in atomic context, annotate driver-ops.h with appropiate might_sleep() calls. At the same time, also document in mac80211.h the op functions with missing contexts. mac80211 doesn't seem to use get_tx_stats anywhere currently. Just to be on the safe side, I documented it to be atomic, but hopefully the op can be removed in the future. Compile-tested only. Signed-off-by:
Kalle Valo <kalle.valo@iki.fi> Signed-off-by:
Johannes Berg <johannes@sipsolutions.net> Signed-off-by:
-
Johannes Berg authored
There's no need to be requeueing the work struct since we check for the scan after removing items due to possible timeouts. Signed-off-by:
-
Johannes Berg authored
All its members (vif, mac_addr, type) are now available in the vif struct directly, so we can pass that instead of the conf struct. I generated this patch (except the mac80211 and header file changes) with this semantic patch: @@ identifier conf, fn, hw; type tp; @@ tp fn(struct ieee80211_hw *hw, -struct ieee80211_if_init_conf *conf) +struct ieee80211_vif *vif) { <... ( -conf->type +vif->type | -conf->mac_addr +vif->addr | -conf->vif +vif ) ...> } Signed-off-by: -
Johannes Berg authored
When, for instance, a new IBSS peer is found, userspace wants to be notified. Add events for all new stations that mac80211 learns about. Signed-off-by:
-
Johannes Berg authored
This changes mac80211 to allow being off-channel for any type of work, not just the 'remain-on-channel' work. This also helps fast transition to a BSS on a different channel. Signed-off-by:
-
Johannes Berg authored
This implements the new remain-on-channel cfg80211 command in mac80211, extending the work interface. Also change the work purge code to be able to clean up events properly (pretending they timed out.) Signed-off-by:
Jouni Malinen <jouni.malinen@atheros.com> Signed-off-by:
-
Jouni Malinen authored
Add new commands for requesting the driver to remain awake on a specified channel for the specified amount of time (and another command to cancel such an operation). This can be used to implement userspace-controlled off-channel operations, like Public Action frame exchange on another channel than the operation channel. The off-channel operation should behave similarly to scan, i.e. the local station (if associated) moves into power save mode to request the AP to buffer frames for it and then moves to the other channel to allow the off-channel operation to be completed. The duration parameter can be used to request enough time to receive a response from the target station. Signed-off-by:
-
Jouni Malinen authored
The off-channel operations for going into power save mode (station mode) or stop beaconing (AP/IBSS) are not limited to scanning. Move these into a separate file and allow them to be used for other purposes, too. Signed-off-by:
-
Johannes Berg authored
cfg80211 offers private data for each BSS struct, which mac80211 uses. However, mac80211 uses internal and external (cfg80211) BSS pointers interchangeably and has a hack to put the cfg80211 bss struct into the private struct. Remove this hack, properly converting between the pointers wherever necessary. Signed-off-by:
-
Johannes Berg authored
Currently, we insert all user-specified IEs before the HT IE for association, and after the HT IE for probe requests. For association, that's correct only if the user-specified IEs are RSN only, incorrect in all other cases including WPA. Change this to split apart the user-specified IEs in two places for association: before the HT IE (e.g. RSN), after the HT IE (generally empty right now I think?) and after WMM (all other vendor-specific IEs). For probes, split the IEs in different places to be correct according to the spec. Signed-off-by:
-
Johannes Berg authored
Refactor the code to reserve an skb of the right size (instead of hoping 200 bytes are enough forever), and also put HT IE generation into an own function. Additionally, put the HT IE before the vendor-specific WMM IE. This still leaves things not quite ordered correctly, due to user-specified IEs, add a note about that for now. Signed-off-by:
-
Johannes Berg authored
The station we're authenticating/associating with may not always be an AP in the sense that word is mostly understood, so print only the MAC address of the peer instead. Signed-off-by:
-
Johannes Berg authored
In order to use auth/assoc for different purposes other than MLME, it needs to be split up. For other purposes, a generic work handling (potentially on another channel) will be useful. To achieve that, this patch moves much of the MLME work handling out of mlme into a new work API. The API can currently handle probing a specific AP, authentication and association. The MLME previously handled probe/authentication as one step and will continue to do so, but they are separate in the new work handling. Work items are RCU-managed to be able to check for existence of an item for a specific frame in the RX path, but they can be re-used which the MLME right now will do for its combined probe/auth step. Signed-off-by:
-
Johannes Berg authored
As a first step of generalising management work, this renames a few things and puts more information directly into the struct so that auth/assoc need not access the BSS pointer as often -- in fact it can be removed from auth completely. Also since the previous patch made sure a new work item is used for association, we can make the different data a union. Signed-off-by:
-
