- 04 May, 2009 8 commits
-
-
Jiri Slaby authored
minstrel doesn't count max rate count in fact, since it doesn't use a loop variable `i' and hence allocs space only for bitrates found in the first band. Fix it by involving the `i' as an index so that it traverses all the bands now and finds the real max bitrate count. Signed-off-by:
Jiri Slaby <jirislaby@gmail.com> Cc: Felix Fietkau <nbd@openwrt.org> Signed-off-by:
John W. Linville <linville@tuxdriver.com>
-
Luis R. Rodriguez authored
Signed-off-by:
Luis R. Rodriguez <lrodriguez@atheros.com> Signed-off-by:
-
Luis R. Rodriguez authored
During initialization we would not have received any beacons so skip processing reg beacon hints, also adds a check to reg_is_world_roaming() for last_request before accessing its fields. This should fix this: BUG: unable to handle kernel NULL pointer dereference at IP: [<e0171332>] wiphy_update_regulatory+0x20f/0x295 *pdpt = 0000000008bf1001 *pde = 0000000000000000 Oops: 0000 [#1] last sysfs file: /sys/class/backlight/eeepc/brightness Modules linked in: ath5k(+) mac80211 led_class cfg80211 go_bit cfbcopyarea cfbimgblt cfbfillrect ipv6 ydev usual_tables(P) snd_hda_codec_realtek snd_hda_intel nd_hwdep uhci_hcd snd_pcm_oss snd_mixer_oss i2c_i801 e serio_raw i2c_core pcspkr atl2 snd_pcm intel_agp re agpgart eeepc_laptop snd_page_alloc ac video backlight rfkill button processor evdev thermal fan ata_generic Pid: 2909, comm: modprobe Tainted: Pc #112) 701 EIP: 0060:[<e0171332>] EFLAGS: 00010246 CPU: 0 EIP is at wiphy_update_regulatory+0x20f/0x295 [cfg80211] EAX: 00000000 EBX: c5da0000 ECX: 00000000 EDX: c5da0060 ESI: 0000001a EDI: c5da0060 EBP: df3bdd70 ESP: df3bdd40 DS: 007b ES: 007b FS: 0000 GS: 0033 SS: 0068 Process modprobe (pid: 2909, ti=df3bc000 task=c5d030000) Stack: df3bdd90 c5da0060 c04277e0 00000001 00000044 c04277e402 00000002 c5da0000 0000001a c5da0060 df3bdda8 e01706a2 02 00000282 000080d0 00000068 c5d53500 00000080 0000028240 Call Trace: [<e01706a2>] ? wiphy_register+0x122/0x1b7 [cfg80211] [<e0328e02>] ? ieee80211_register_hw+0xd8/0x346 [<e06a7c9f>] ? ath5k_hw_set_bssid_mask+0x71/0x78 [ath5k] [<e06b0c52>] ? ath5k_pci_probe+0xa5c/0xd0a [ath5k] [<c01a6037>] ? sysfs_find_dirent+0x16/0x27 [<c01fec95>] ? local_pci_probe+0xe/0x10 [<c01ff526>] ? pci_device_probe+0x48/0x66 [<c024c9fd>] ? driver_probe_device+0x7f/0xf2 [<c024cab3>] ? __driver_attach+0x43/0x5f [<c024c0af>] ? bus_for_each_dev+0x39/0x5a [<c024c8d0>] ? driver_attach+0x14/0x16 [<c024ca70>] ? __driver_attach+0x0/0x5f [<c024c5b3>] ? bus_add_driver+0xd7/0x1e7 [<c024ccb9>] ? driver_register+0x7b/0xd7 [<c01ff827>] ? __pci_register_driver+0x32/0x85 [<e00a8018>] ? init_ath5k_pci+0x18/0x30 [ath5k] [<c0101131>] ? _stext+0x49/0x10b [<e00a8000>] ? init_ath5k_pci+0x0/0x30 [ath5k] [<c012f452>] ? __blocking_notifier_call_chain+0x40/0x4c [<c013a714>] ? sys_init_module+0x87/0x18b [<c0102804>] ? sysenter_do_call+0x12/0x22 Code: b8 da 17 e0 83 c0 04 e8 92 f9 ff ff 84 c0 75 2a 8b 85 c0 74 0c 83 c0 04 e8 7c f9 ff ff 84 c0 75 14 a1 bc da 4 03 74 66 8b 4d d4 80 79 08 00 74 5d a1 e0 d2 17 e0 48 EIP: [<e0171332>] wiphy_update_regulatory+0x20f/0x295 SP 0068:df3bdd40 CR2: 0000000000000004 ---[ end trace 830f2dd2a95fd1a8 ]--- This issue is hard to reproduce, but it was noticed and discussed on this thread: http://marc.info/?t=123938022700005&r=1&w=2 Cc: stable@kernel.org Reported-by:
Alan Jenkins <alan-jenkins@tuffmail.co.uk> Signed-off-by:
-
Luis R. Rodriguez authored
We forgot to lock using the cfg80211_mutex in wiphy_apply_custom_regulatory(). Without the lock there is possible race between processing a reply from CRDA and a driver calling wiphy_apply_custom_regulatory(). During the processing of the reply from CRDA we free last_request and wiphy_apply_custom_regulatory() eventually accesses an element from last_request in the through freq_reg_info_regd(). This is very difficult to reproduce (I haven't), it takes us 3 hours and you need to be banging hard, but the race is obvious by looking at the code. This should only affect those who use this caller, which currently is ath5k, ath9k, and ar9170. EIP: 0060:[<f8ebec50>] EFLAGS: 00210282 CPU: 1 EIP is at freq_reg_info_regd+0x24/0x121 [cfg80211] EAX: 00000000 EBX: f7ca0060 ECX: f5183d94 EDX: 0024cde0 ESI: f8f56edc EDI: 00000000 EBP: 00000000 ESP: f5183d44 DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068 Process modprobe (pid: 14617, ti=f5182000 task=f3934d10 task.ti=f5182000) Stack: c0505300 f7ca0ab4 f5183d94 0024cde0 f8f403a6 f8f63160 f7ca0060 00000000 00000000 f8ebedf8 f5183d90 f8f56edc 00000000 00000004 00000f40 f8f56edc f7ca0060 f7ca1234 00000000 00000000 00000000 f7ca14f0 f7ca0ab4 f7ca1289 Call Trace: [<f8ebedf8>] wiphy_apply_custom_regulatory+0x8f/0x122 [cfg80211] [<f8f3f798>] ath_attach+0x707/0x9e6 [ath9k] [<f8f45e46>] ath_pci_probe+0x18d/0x29a [ath9k] [<c023c7ba>] pci_device_probe+0xa3/0xe4 [<c02a860b>] really_probe+0xd7/0x1de [<c02a87e7>] __driver_attach+0x37/0x55 [<c02a7eed>] bus_for_each_dev+0x31/0x57 [<c02a83bd>] driver_attach+0x16/0x18 [<c02a78e6>] bus_add_driver+0xec/0x21b [<c02a8959>] driver_register+0x85/0xe2 [<c023c9bb>] __pci_register_driver+0x3c/0x69 [<f8e93043>] ath9k_init+0x43/0x68 [ath9k] [<c010112b>] _stext+0x3b/0x116 [<c014a872>] sys_init_module+0x8a/0x19e [<c01049ad>] sysenter_do_call+0x12/0x21 [<ffffe430>] 0xffffe430 ======================= Code: 0f 94 c0 c3 31 c0 c3 55 57 56 53 89 c3 83 ec 14 8b 74 24 2c 89 54 24 0c 89 4c 24 08 85 f6 75 06 8b 35 c8 bb ec f8 a1 cc bb ec f8 <8b> 40 04 83 f8 03 74 3a 48 74 37 8b 43 28 85 c0 74 30 89 c6 8b EIP: [<f8ebec50>] freq_reg_info_regd+0x24/0x121 [cfg80211] SS:ESP 0068:f5183d44 Cc: stable@kernel.org Reported-by:
Nataraj Sadasivam <Nataraj.Sadasivam@Atheros.com> Reported-by:
Vivek Natarajan <Vivek.Natarajan@Atheros.com> Signed-off-by:
-
Reinette Chatre authored
We need to be symmetrical in what is done when key is set and cleared. This is important wrt the key flags as they are used during key clearing and if they are not set when the key is set the key cannot be cleared completely. This addresses the many occurences of the WARN found in iwl_set_tkip_dynamic_key_info() and tracked in http://www.kerneloops.org/searchweek.php?search=iwl_set_dynamic_key If calling iwl_set_tkip_dynamic_key_info()/iwl_remove_dynamic_key() pair a few times in a row will cause that we run out of key space. This is because the index stored in the key flags is used by iwl_remove_dynamic_key() to decide if it should remove the key. Unfortunately the key flags, and hence the key index is currently only set at the time the key is written to the device (in iwl_update_tkip_key()) and _not_ in iwl_set_tkip_dynamic_key_info(). Fix this by setting flags in iwl_set_tkip_dynamic_key_info(). Signed-off-by:
Reinette Chatre <reinette.chatre@intel.com> Signed-off-by:
-
Johannes Berg authored
Another bug in the "cfg80211: do not replace BSS structs" patch, a forgotten length update leads to bogus data being stored and passed to userspace, often truncated. Signed-off-by:
Johannes Berg <johannes@sipsolutions.net> Signed-off-by:
-
Johannes Berg authored
The fragmentation threshold is defined to be including the FCS, and the code that sets the TX_FRAGMENTED flag correctly accounts for those four bytes. The code that verifies this doesn't though, which could lead to spurious warnings and frames being dropped although everything is ok. Correct the code by accounting for the FCS. (JWL -- The problem is described here: http://article.gmane.org/gmane.linux.kernel.wireless.general/32205 ) Signed-off-by:
-
Andreas Schwab authored
It does not make sense to apply EXPORT_SYMBOL to a static symbol. Fixes this build error: drivers/net/wireless/iwlwifi/iwl3945-base.c:1697: error: __ksymtab_iwl3945_rx_queue_reset causes a section type conflict Signed-off-by:
Andreas Schwab <schwab@linux-m68k.org> Signed-off-by:
-
- 02 May, 2009 5 commits
-
-
Lubomir Rintel authored
Doing it in reverse order causes uevent to be sent before we have a MAC address, which confuses udev. Signed-off-by:
Lubomir Rintel <lkundrak@v3.sk> Acked-by:
Jeff Garzik <jgarzik@redhat.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Rabin Vincent authored
Commit 0ba25ff4 ("br2684: convert to net_device_ops") inadvertently deleted the initialization of the net_dev pointer in the br2684_dev structure, leading to crashes. This patch adds it back. Reported-by:
Mikko Vinni <mmvinni@yahoo.com> Tested-by:
Rabin Vincent <rabin@rab.in> Signed-off-by:
-
Robert Love authored
The kernel should only be using the high 16 bits of a kernel generated priority. Filter priorities in all other cases only use the upper 16 bits of the u32 'prio' field of 'struct tcf_proto', but when the kernel generates the priority of a filter is saves all 32 bits which can result in incorrect lookup failures when a filter needs to be deleted or modified. Signed-off-by:
Robert Love <robert.w.love@intel.com> Signed-off-by:
-
Alex Williamson authored
Signed-off-by:
Alex Williamson <alex.williamson@hp.com> Signed-off-by:
-
Alex Williamson authored
We were avoiding calling sg_init* on scatterlists passed into virtnet_send_command to prevent extraneous end markers. This caused build warnings for uninitialized variables. Cleanup the code to create proper scatterlists. Signed-off-by:
-
- 01 May, 2009 10 commits
-
-
Jiri Pirko authored
This patch makes the cleanup in bond_create nicer :) Also now the forgotten free_netdev is called. Signed-off-by:
Jiri Pirko <jpirko@redhat.com> Signed-off-by:
-
Grant Likely authored
virtio_net.h uses the macro ETH_ALEN which is defined in linux/if_ether.h. Discovered when hacking on virtio-over-pci patches. Signed-off-by:
Grant Likely <grant.likely@secretlab.ca> Signed-off-by:
-
Steve Glendinning authored
LAN9512 and LAN9514 are USB hubs with an integrated 10/100 ethernet controller. Logically this looks like an ethernet controller (similar to LAN9500) permanently attached to one of the hub's downstream ports. This patch adds the usb device id of the new ethernet controller to the smsc95xx driver. This id is the same in both new devices. Signed-off-by:
Steve Glendinning <steve.glendinning@smsc.com> Signed-off-by:
-
Steve Glendinning authored
SMSC LAN9500 has dual purpose GPIO/LED pins, and by default at power-on these are configured as GPIOs. This means that if LEDs are fitted they won't ever light. This patch sets them to be LED outputs for speed, duplex and link/activity. Signed-off-by:
-
Bruno Prémont authored
When netconsole is loaded and a network interface fades away (e.g. on rmmod $interface_driver_module) the rmmod remains stuck and some locks are taken that prevent any additional module loading/unloading as well as interface up/down changes. In addition kernel logs (and console) get flooded at 10s interval with [ 122.464065] unregister_netdevice: waiting for eth0 to become free. Usage count = 1 [ 132.704059] unregister_netdevice: waiting for eth0 to become free. Usage count = 1 This patch lets netconsole take NETDEV_UNREGISTER event into account and release the affected interface if it was in use. Signed-off-by:
Bruno Prémont <bonbons@linux-vserver.org> Acked-by:
Matt Mackall <mpm@selenic.com> Signed-off-by:
-
Laszlo Attila Toth authored
xt_socket can use connection tracking, and checks whether it is a module. Signed-off-by:
Laszlo Attila Toth <panther@balabit.hu> Signed-off-by:
-
Eric Dumazet authored
bond_slave_info_query() should keep a read lock while accessing slave info, or risk accessing stale data and corruption. Signed-off-by:
Eric Dumazet <dada1@cosmosbay.com> Signed-off-by:
Jay Vosburgh <fubar@us.ibm.com> Signed-off-by:
-
Sergey Senozhatsky authored
Trivial: fixing gcc 4.4 compiler warning: drivers/net/cxgb3/t3_hw.c: In function ‘t3_prep_adapter’: drivers/net/cxgb3/t3_hw.c:3782: warning: suggest parentheses around operand of ‘!’ or change ‘|’ to ‘||’ or ‘!’ to ‘~’ Signed-off-by:
Sergey Senozhatsky <sergey.senozhatsky@mail.by> Signed-off-by:
-
Eric Dumazet authored
Signed-off-by:
-
Eric Dumazet authored
When skb_rx_queue_recorded() is true, we dont want to use jash distribution as the device driver exactly told us which queue was selected at RX time. jhash makes a statistical shuffle, but this wont work with 8 static inputs. Later improvements would be to compute reciprocal value of real_num_tx_queues to avoid a divide here. But this computation should be done once, when real_num_tx_queues is set. This needs a separate patch, and a new field in struct net_device. Reported-by:
Andrew Dickinson <andrew@whydna.net> Signed-off-by:
-
- 30 Apr, 2009 3 commits
-
-
Jarek Poplawski authored
Lennert Buytenhek wrote: > Since 4fb66994 ("net: Optimize memory > usage when splicing from sockets.") I'm seeing this oops (e.g. in > 2.6.30-rc3) when splicing from a TCP socket to /dev/null on a driver > (mv643xx_eth) that uses LRO in the skb mode (lro_receive_skb) rather > than the frag mode: My patch incorrectly assumed skb->sk was always valid, but for "frag_listed" skbs we can only use skb->sk of their parent. Reported-by:
Lennert Buytenhek <buytenh@wantstofly.org> Debugged-by:
Jarek Poplawski <jarkao2@gmail.com> Signed-off-by:
-
Lennert Buytenhek authored
On several mv643xx_eth hardware versions, the two 64bit mib counters for 'good octets received' and 'good octets sent' are actually 32bit counters, and reading from the upper half of the register has the same effect as reading from the lower half of the register: an atomic read-and-clear of the entire 32bit counter value. This can under heavy traffic occasionally lead to small numbers being added to the upper half of the 64bit mib counter even though no 32bit wrap has occured. Since we poll the mib counters at least every 30 seconds anyway, we might as well just skip the reads of the upper halves of the hardware counters without breaking the stats, which this patch does. Signed-off-by:
Lennert Buytenhek <buytenh@marvell.com> Cc: stable@kernel.org Signed-off-by:
-
Lennert Buytenhek authored
Currently, when OOM occurs during rx ring refill, mv643xx_eth will get into an infinite loop, due to the refill function setting the OOM bit but not clearing the 'rx refill needed' bit for this queue, while the calling function (the NAPI poll handler) will call the refill function in a loop until the 'rx refill needed' bit goes off, without checking the OOM bit. This patch fixes this by checking the OOM bit in the NAPI poll handler before attempting to do rx refill. This means that once OOM occurs, we won't try to do any memory allocations again until the next invocation of the poll handler. While we're at it, change the OOM flag to be a single bit instead of one bit per receive queue since OOM is a system state rather than a per-queue state, and cancel the OOM timer on entry to the NAPI poll handler if it's running to prevent it from firing when we've already come out of OOM. Signed-off-by:
-
- 29 Apr, 2009 6 commits
-
-
-
Johannes Berg authored
In "mac80211: correct wext transmit power handler" I fixed the wext handler, but forgot to make the default of the user_power_level -1 (aka "auto"), so that now the transmit power is always set to 0, causing associations to time out and similar problems since we're transmitting with very little power. Correct this by correcting the default user_power_level to -1. Signed-off-by:
Niel Lambrechts <niel.lambrechts@gmail.com> Signed-off-by:
-
Alan Jenkins authored
- ieee80211_wep_init(), which is called with rtnl_lock held, blocks in request_module() [waiting for modprobe to load a crypto module]. - modprobe blocks in a call to flush_workqueue(), when it closes a TTY [presumably when it exits]. - The workqueue item linkwatch_event() blocks on rtnl_lock. There's no reason for wep_init() to be called with rtnl_lock held, so just move it outside the critical section. Signed-off-by:
-
Thadeu Lima de Souza Cascardo authored
After experimenting with kexec with the last merges after 2.6.29, I've had some problems when probing e100. It would not read the eeprom. After some bisects, I realized this has been like that since forever (at least 2.6.18). The problem is that shutdown is doing the same thing that suspend does and puts the device in D3 state. I couldn't find a way to get the device back to a sane state in the probe function. So, based on some similar patches from Rafael J. Wysocki for e1000, e1000e, and ixgbe, I wrote this one for e100. Signed-off-by:
Thadeu Lima de Souza Cascardo <cascardo@holoscopio.com> Acked-by:
"Rafael J. Wysocki" <rjw@sisk.pl> Signed-off-by:
Jeff Kirsher <jeffrey.t.kirsher@intel.com> Signed-off-by:
-
-
Stephen Hemminger authored
The x_tables are organized with a table structure and a per-cpu copies of the counters and rules. On older kernels there was a reader/writer lock per table which was a performance bottleneck. In 2.6.30-rc, this was converted to use RCU and the counters/rules which solved the performance problems for do_table but made replacing rules much slower because of the necessary RCU grace period. This version uses a per-cpu set of spinlocks and counters to allow to table processing to proceed without the cache thrashing of a global reader lock and keeps the same performance for table updates. Signed-off-by:
Stephen Hemminger <shemminger@vyatta.com> Acked-by:
Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by:
-
- 28 Apr, 2009 8 commits
-
-
Bob Copeland authored
char bname[5] is too small for the string "X GHz" when the null terminator is taken into account. Thus, turning on rate debugging can crash unless we have lucky stack alignment. Cc: stable@kernel.org Reported-by:
Paride Legovini <legovini@spiro.fisica.unipd.it> Signed-off-by:
Bob Copeland <me@bobcopeland.com> Signed-off-by:
-
Johannes Berg authored
Under certain circumstances iwlwifi can get stuck and will no longer accept scan requests, because the core code (cfg80211) thinks that it's still processing one. This fixes one of the points where it can happen, but I've still seen it (although only with my radio-off-when-idle patch). Signed-off-by:
-
Jussi Kivilinna authored
rndis_wext_link_change() might be called from rndis_command() at initialization stage and priv->workqueue/priv->work have not been initialized yet. This causes invalid opcode at rndis_wext_bind on some brands of bcm4320. Fix by initializing workqueue/workers in rndis_wext_bind() before rndis_command is used. This bug has existed since 2.6.25, reported at: http://bugzilla.kernel.org/show_bug.cgi?id=12794Signed-off-by:
Jussi Kivilinna <jussi.kivilinna@mbnet.fi> Signed-off-by:
-
Stephen Rothwell authored
drivers/net/wireless/iwlwifi/iwl3945-base.c:1415: error: __ksymtab_iwl3945_rx_queue_reset causes a section type conflict I am pretty sure that this is a compiler bug, so not to worry. However, as far as I can see, iwl-3945.o (the only user) and iwl3945-base.o are always linked into the same module, so the EXPORT_SYMBOL (which causes the problem) should not be needed. Correct? Signed-off-by:
Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by:
-
Marcel Holtmann authored
The Bluetooth 2.1 specification introduced four different security modes that can be mapped using Legacy Pairing and Simple Pairing. With the usage of Simple Pairing it is required that all connections (except the ones for SDP) are encrypted. So even the low security requirement mandates an encrypted connection when using Simple Pairing. When using Legacy Pairing (for Bluetooth 2.0 devices and older) this is not required since it causes interoperability issues. To support this properly the low security requirement translates into different host controller transactions depending if Simple Pairing is supported or not. However in case of Simple Pairing the command to switch on encryption after a successful authentication is not triggered for the low security mode. This patch fixes this and actually makes the logic to differentiate between Simple Pairing and Legacy Pairing a lot simpler. Based on a report by Ville Tervo <ville.tervo@nokia.com> Signed-off-by:Marcel Holtmann <marcel@holtmann.org>
-
Marcel Holtmann authored
The Bluetooth stack uses a reference counting for all established ACL links and if no user (L2CAP connection) is present, the link will be terminated to save power. The problem part is the dedicated pairing when using Legacy Pairing (Bluetooth 2.0 and before). At that point no user is present and pairing attempts will be disconnected within 10 seconds or less. In previous kernel version this was not a problem since the disconnect timeout wasn't triggered on incoming connections for the first time. However this caused issues with broken host stacks that kept the connections around after dedicated pairing. When the support for Simple Pairing got added, the link establishment procedure needed to be changed and now causes issues when using Legacy Pairing When using Simple Pairing it is possible to do a proper reference counting of ACL link users. With Legacy Pairing this is not possible since the specification is unclear in some areas and too many broken Bluetooth devices have already been deployed. So instead of trying to deal with all the broken devices, a special pairing timeout will be introduced that increases the timeout to 60 seconds when pairing is triggered. If a broken devices now puts the stack into an unforeseen state, the worst that happens is the disconnect timeout triggers after 120 seconds instead of 4 seconds. This allows successful pairings with legacy and broken devices now. Based on a report by Johan Hedberg <johan.hedberg@nokia.com> Signed-off-by: -
Roger Quadros authored
Use a different work_struct variables for add_conn() and del_conn() and use single work queue instead of two for adding and deleting connections. It eliminates the following error on a preemptible kernel: [ 204.358032] Unable to handle kernel NULL pointer dereference at virtual address 0000000c [ 204.370697] pgd = c0004000 [ 204.373443] [0000000c] *pgd=00000000 [ 204.378601] Internal error: Oops: 17 [#1] PREEMPT [ 204.383361] Modules linked in: vfat fat rfcomm sco l2cap sd_mod scsi_mod iphb pvr2d drm omaplfb ps [ 204.438537] CPU: 0 Not tainted (2.6.28-maemo2 #1) [ 204.443664] PC is at klist_put+0x2c/0xb4 [ 204.447601] LR is at klist_put+0x18/0xb4 [ 204.451568] pc : [<c0270f08>] lr : [<c0270ef4>] psr: a0000113 [ 204.451568] sp : cf1b3f10 ip : cf1b3f10 fp : cf1b3f2c [ 204.463104] r10: 00000000 r9 : 00000000 r8 : bf08029c [ 204.468353] r7 : c7869200 r6 : cfbe2690 r5 : c78692c8 r4 : 00000001 [ 204.474945] r3 : 00000001 r2 : cf1b2000 r1 : 00000001 r0 : 00000000 [ 204.481506] Flags: NzCv IRQs on FIQs on Mode SVC_32 ISA ARM Segment kernel [ 204.488861] Control: 10c5387d Table: 887fc018 DAC: 00000017 [ 204.494628] Process btdelconn (pid: 515, stack limit = 0xcf1b22e0) Signed-off-by:
Roger Quadros <ext-roger.quadros@nokia.com> Signed-off-by:
-
Eric Dumazet authored
In 2.6.25 we added UDP mem accounting. This unfortunatly added a penalty when a frame is transmitted, since we have at TX completion time to call sock_wfree() to perform necessary memory accounting. This calls sock_def_write_space() and utimately scheduler if any thread is waiting on the socket. Thread(s) waiting for an incoming frame was scheduled, then had to sleep again as event was meaningless. (All threads waiting on a socket are using same sk_sleep anchor) This adds lot of extra wakeups and increases latencies, as noted by Christoph Lameter, and slows down softirq handler. Reference : http://marc.info/?l=linux-netdev&m=124060437012283&w=2 Fortunatly, Davide Libenzi recently added concept of keyed wakeups into kernel, and particularly for sockets (see commit 37e5540b epoll keyed wakeups: make sockets use keyed wakeups) Davide goal was to optimize epoll, but this new wakeup infrastructure can help non epoll users as well, if they care to setup an appropriate handler. This patch introduces new DEFINE_WAIT_FUNC() helper and uses it in wait_for_packet(), so that only relevant event can wakeup a thread blocked in this function. Trace of function calls from bnx2 TX completion bnx2_poll_work() is : __kfree_skb() skb_release_head_state() sock_wfree() sock_def_write_space() __wake_up_sync_key() __wake_up_common() receiver_wake_function() : Stops here since thread is waiting for an INPUT Reported-by:
Christoph Lameter <cl@linux.com> Signed-off-by:
-
