- 03 Mar, 2015 3 commits
-
-
Ben Hutchings authored
This reverts commit fd9af07c. The hardware manual states that the frame error and multicast bits are copied to bits 9:0 of RD0, not bits 25:16. I've tested that this is true for RFS1 (CRC error), RFS3 (frame too short), RFS4 (frame too long) and RFS8 (multicast). Also adjust a comment to agree with this. Signed-off-by:
Ben Hutchings <ben.hutchings@codethink.co.uk> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Ben Hutchings authored
In case of RX ring underrun (RDE), we attempt to reset the software descriptor pointers (dirty_rx and cur_rx) to match where the hardware will read the next descriptor from, as that might not be the first dirty descriptor. This relies on reading RDFAR, but that register doesn't exist on all supported chips - specifically, not on the R-Car chips. This will result in unpredictable behaviour on those chips after an RDE. Make this pointer reset conditional and assume that it isn't needed on the R-Car chips. This fix also assumes that RDFAR is never exposed at offset 0 in the memory map - this is currently true, and a subsequent commit will fix the ambiguity between offset 0 and no-offset in the register offset maps. Fixes: 79fba9f5 ("net: sh_eth: fix the rxdesc pointer when rx ...") Signed-off-by:
-
Ben Hutchings authored
When submitting a DMA descriptor, the active bit must be written last. When reading a completed DMA descriptor, the active bit must be read first. Add memory barriers to ensure that this ordering is maintained. Signed-off-by:
-
- 02 Mar, 2015 3 commits
-
-
David S. Miller authored
Or Gerlitz says: ==================== Mellanox driver fixes Two small fixes, please apply to net. Both patches should go to 3.19.y too. ==================== Signed-off-by: -
Ido Shamay authored
Packets which are sent from the selftest (ethtool) flow, should not be passed to GRO stack but rather dropped by the driver after validation. To achieve that, we disable GRO for the duration of the selftest. Fixes: dd65beac ("net/mlx4_en: Extend usage of napi_gro_frags") Reported-by:
Carol Soto <clsoto@linux.vnet.ibm.com> Signed-off-by:
Ido Shamay <idos@mellanox.com> Signed-off-by:
Or Gerlitz <ogerlitz@mellanox.com> Signed-off-by:
-
Or Gerlitz authored
The bit mask for currently supported driver features (MLX4_UPDATE_QP_SUPPORTED_ATTRS) of the update-qp command was defined twice (using enum value and pre-processor define directive) and wrong. The return value of the call to mlx4_update_qp() from within the SRIOV resource-tracker was wrongly voided down. Fix both issues. issue: none Fixes: 09e05c3f ('net/mlx4: Set vlan stripping policy by the right command') Fixes: ce8d9e0d ('net/mlx4_core: Add UPDATE_QP SRIOV wrapper support') Signed-off-by:
Matan Barak <matanb@mellanox.com> Signed-off-by:
-
- 01 Mar, 2015 20 commits
-
-
David S. Miller authored
Florian Fainelli says: ==================== net: bcmgenet and systemport statistics fixes This two patches fix a similar problem in the GENET and SYSTEMPORT drivers for software maintained statistics used to track DMA mapping and SKB re-allocation failures. ==================== Signed-off-by: -
Florian Fainelli authored
Commit 60b4ea17 ("net: systemport: log RX buffer allocation and RX/TX DMA failures") added a few software maintained statistics using BCM_SYSPORT_STAT_MIB_RX and BCM_SYSPORT_STAT_MIB_TX. These statistics are read from the hardware MIB counters, such that bcm_sysport_update_mib_counters() was trying to read from a non-existing MIB offset for these counters. Fix this by introducing a special type: BCM_SYSPORT_STAT_SOFT, similar to BCM_SYSPORT_STAT_NETDEV, such that bcm_sysport_get_ethtool_stats will read from the software mib. Fixes: 60b4ea17 ("net: systemport: log RX buffer allocation and RX/TX DMA failures") Signed-off-by:
Florian Fainelli <f.fainelli@gmail.com> Signed-off-by:
-
Florian Fainelli authored
Commit 44c8bc3c ("net: bcmgenet: log RX buffer allocation and RX/TX dma failures") added a few software maintained statistics using BCMGENET_STAT_MIB_RX and BCMGENET_STAT_MIB_TX. These statistics are read from the hardware MIB counters, such that bcmgenet_update_mib_counters() was trying to read from a non-existing MIB offset for these counters. Fix this by introducing a special type: BCMGENET_STAT_SOFT, similar to BCMGENET_STAT_NETDEV, such that bcmgenet_get_ethtool_stats will read from the software mib. Fixes: 44c8bc3c ("net: bcmgenet: log RX buffer allocation and RX/TX dma failures") Signed-off-by:
-
Florian Westphal authored
rxrpc_resend_timeout has an initial value of 4 * HZ; use it as-is. Signed-off-by:
Florian Westphal <fw@strlen.de> Signed-off-by:
-
Florian Westphal authored
Typo, 'stop' is never set to true. Seems intent is to not attempt to retransmit more packets after sendmsg returns an error. This change is based on code inspection only. Signed-off-by:
-
Arvid Brodin authored
To repeat: $ sudo ip link del hsr0 BUG: unable to handle kernel NULL pointer dereference at 0000000000000018 IP: [<ffffffff8187f495>] hsr_del_port+0x15/0xa0 etc... Bug description: As part of the hsr master device destruction, hsr_del_port() is called for each of the hsr ports. At each such call, the master device is updated regarding features and mtu. When the master device is freed before the slave interfaces, master will be NULL in hsr_del_port(), which led to a NULL pointer dereference. Additionally, dev_put() was called on the master device itself in hsr_del_port(), causing a refcnt error. A third bug in the same code path was that the rtnl lock was not taken before hsr_del_port() was called as part of hsr_dev_destroy(). The reporter (Nicolas Dichtel) also said: "hsr_netdev_notify() supposes that the port will always be available when the notification is for an hsr interface. It's wrong. For example, netdev_wait_allrefs() may resend NETDEV_UNREGISTER.". As a precaution against this, a check for port == NULL was added in hsr_dev_notify(). Reported-by:
Nicolas Dichtel <nicolas.dichtel@6wind.com> Fixes: 51f3c605 ("net/hsr: Move slave init to hsr_slave.c.") Signed-off-by:
Arvid Brodin <arvid.brodin@alten.se> Signed-off-by:
-
Vaishali Thakkar authored
Use timer API functions setup_timer and mod_timer instead of structure assignments as they are standard way to set the timer and to update the expire field of an active timer respectively. This is done using Coccinelle and semantic patch used for this is as follows: // <smpl> @@ expression x,y,z,a,b; @@ -init_timer (&x); +setup_timer (&x, y, z); +mod_timer (&a, b); -x.function = y; -x.data = z; -x.expires = b; -add_timer(&a); // </smpl> Signed-off-by:
Vaishali Thakkar <vthakkar1994@gmail.com> Signed-off-by:
-
Vaishali Thakkar authored
Use timer API functions setup_timer and mod_timer instead of structure assignments as they are standard way to set the timer and to update the expire field of an active timer respectively. This is done using Coccinelle and semantic patch used for this is as follows: // <smpl> @@ expression x,y,z,a,b; @@ -init_timer (&x); +setup_timer (&x, y, z); +mod_timer (&a, b); -x.function = y; -x.data = z; -x.expires = b; -add_timer(&a); // </smpl> Signed-off-by:
-
Vaishali Thakkar authored
Use timer API functions setup_timer and mod_timer instead of structure assignments as they are standard way to set the timer and to update the expire field of an active timer respectively. This is done using Coccinelle and semantic patch used for this is as follows: // <smpl> @@ expression x,y,z,a,b; @@ -init_timer (&x); +setup_timer (&x, y, z); +mod_timer (&a, b); -x.function = y; -x.data = z; -x.expires = b; -add_timer(&a); // </smpl> Signed-off-by:
-
Vaishali Thakkar authored
Use timer API functions setup_timer and mod_timer instead of structure assignments as they are standard way to set the timer and to update the expire field of an active timer respectively. This is done using Coccinelle and semantic patch used for this is as follows: // <smpl> @@ expression x,y,z,a,b; @@ -init_timer (&x); +setup_timer (&x, y, z); +mod_timer (&a, b); -x.function = y; -x.data = z; -x.expires = b; -add_timer(&a); // </smpl> Signed-off-by:
-
Vaishali Thakkar authored
Use timer API functions setup_timer and mod_timer instead of structure assignments as they are standard way to set the timer and to update the expire field of an active timer respectively. This is done using Coccinelle and semantic patch used for this is as follows: // <smpl> @@ expression x,y,z,a,b; @@ -init_timer (&x); +setup_timer (&x, y, z); +mod_timer (&a, b); -x.function = y; -x.data = z; -x.expires = b; -add_timer(&a); // </smpl> Signed-off-by:
-
Yannick Guerrini authored
Change 'mutliple' to 'multiple' Change 'Firmare' to 'Firmware' Signed-off-by:
Yannick Guerrini <yguerrini@tomshardware.fr> Signed-off-by:
-
Yannick Guerrini authored
Change 'Firmare' to 'Firmware' Signed-off-by:
-
Grygorii Strashko authored
Setting a dev_pm_ops suspend/resume pair but not a set of hibernation functions means those pm functions will not be called upon hibernation. Fix this by using SIMPLE_DEV_PM_OPS, which appropriately assigns the suspend and hibernation handlers and move cpsw_suspend/resume calbacks under CONFIG_PM_SLEEP to avoid build warnings. Signed-off-by:
Grygorii Strashko <Grygorii.Strashko@linaro.org> Signed-off-by:
-
Grygorii Strashko authored
Setting a dev_pm_ops suspend_late/resume_early pair but not a set of hibernation functions means those pm functions will not be called upon hibernation. Fix this by using SET_LATE_SYSTEM_SLEEP_PM_OPS, which appropriately assigns the suspend and hibernation handlers and move davinci_mdio_x callbacks under CONFIG_PM_SLEEP to avoid build warnings. Signed-off-by:
-
Eric Dumazet authored
We did a failed attempt in the past to only use rcu in rtnl dump operations (commit e67f88dd "net: dont hold rtnl mutex during netlink dump callbacks") Now that dumps are holding RTNL anyway, there is no need to also use rcu locking, as it forbids any scheduling ability, like GFP_KERNEL allocations that controlling path should use instead of GFP_ATOMIC whenever possible. This should fix following splat Cong Wang reported : [ INFO: suspicious RCU usage. ] 3.19.0+ #805 Tainted: G W include/linux/rcupdate.h:538 Illegal context switch in RCU read-side critical section! other info that might help us debug this: rcu_scheduler_active = 1, debug_locks = 0 2 locks held by ip/771: #0: (rtnl_mutex){+.+.+.}, at: [<ffffffff8182b8f4>] netlink_dump+0x21/0x26c #1: (rcu_read_lock){......}, at: [<ffffffff817d785b>] rcu_read_lock+0x0/0x6e stack backtrace: CPU: 3 PID: 771 Comm: ip Tainted: G W 3.19.0+ #805 Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011 0000000000000001 ffff8800d51e7718 ffffffff81a27457 0000000029e729e6 ffff8800d6108000 ffff8800d51e7748 ffffffff810b539b ffffffff820013dd 00000000000001c8 0000000000000000 ffff8800d7448088 ffff8800d51e7758 Call Trace: [<ffffffff81a27457>] dump_stack+0x4c/0x65 [<ffffffff810b539b>] lockdep_rcu_suspicious+0x107/0x110 [<ffffffff8109796f>] rcu_preempt_sleep_check+0x45/0x47 [<ffffffff8109e457>] ___might_sleep+0x1d/0x1cb [<ffffffff8109e67d>] __might_sleep+0x78/0x80 [<ffffffff814b9b1f>] idr_alloc+0x45/0xd1 [<ffffffff810cb7ab>] ? rcu_read_lock_held+0x3b/0x3d [<ffffffff814b9f9d>] ? idr_for_each+0x53/0x101 [<ffffffff817c1383>] alloc_netid+0x61/0x69 [<ffffffff817c14c3>] __peernet2id+0x79/0x8d [<ffffffff817c1ab7>] peernet2id+0x13/0x1f [<ffffffff817d8673>] rtnl_fill_ifinfo+0xa8d/0xc20 [<ffffffff810b17d9>] ? __lock_is_held+0x39/0x52 [<ffffffff817d894f>] rtnl_dump_ifinfo+0x149/0x213 [<ffffffff8182b9c2>] netlink_dump+0xef/0x26c [<ffffffff8182bcba>] netlink_recvmsg+0x17b/0x2c5 [<ffffffff817b0adc>] __sock_recvmsg+0x4e/0x59 [<ffffffff817b1b40>] sock_recvmsg+0x3f/0x51 [<ffffffff817b1f9a>] ___sys_recvmsg+0xf6/0x1d9 [<ffffffff8115dc67>] ? handle_pte_fault+0x6e1/0xd3d [<ffffffff8100a3a0>] ? native_sched_clock+0x35/0x37 [<ffffffff8109f45b>] ? sched_clock_local+0x12/0x72 [<ffffffff8109f6ac>] ? sched_clock_cpu+0x9e/0xb7 [<ffffffff810cb7ab>] ? rcu_read_lock_held+0x3b/0x3d [<ffffffff811abde8>] ? __fcheck_files+0x4c/0x58 [<ffffffff811ac556>] ? __fget_light+0x2d/0x52 [<ffffffff817b376f>] __sys_recvmsg+0x42/0x60 [<ffffffff817b379f>] SyS_recvmsg+0x12/0x1c Signed-off-by:
Eric Dumazet <edumazet@google.com> Fixes: 0c7aecd4 ("netns: add rtnl cmd to add and get peer netns ids") Cc: Nicolas Dichtel <nicolas.dichtel@6wind.com> Reported-by:
Cong Wang <xiyou.wangcong@gmail.com> Signed-off-by:
-
Geert Uytterhoeven authored
Commit 740c7f31 ("sh_eth: Ensure DMA engines are stopped before freeing buffers") added a call to sh_eth_reset() to the sh_eth_set_ringparam() and sh_eth_close() paths. However, setting the software reset bit(s) in the EDMR register resets the MAC Address Registers to zero. Hence after kexec, the new kernel doesn't detect a valid MAC address and assigns a random MAC address, breaking DHCP. Set the MAC address again after the reset in sh_eth_dev_exit() to fix this. Tested on r8a7740/armadillo (GETHER) and r8a7791/koelsch (FAST_RCAR). Fixes: 740c7f31 ("sh_eth: Ensure DMA engines are stopped before freeing buffers") Signed-off-by:
Geert Uytterhoeven <geert+renesas@glider.be> Signed-off-by:
-
Jaedon Shin authored
This patch adds bcmgenet_tx_poll for the tx_rings. This can reduce the interrupt load and send xmit in network stack on time. This also separated for the completion of tx_ring16 from bcmgenet_poll. The bcmgenet_tx_reclaim of tx_ring[{0,1,2,3}] operative by an interrupt is to be not more than a certain number TxBDs. It is caused by too slowly reclaiming the transmitted skb. Therefore, performance degradation of xmit after 605ad7f1 ("tcp: refine TSO autosizing"). Signed-off-by:Jaedon Shin <jaedon.shin@gmail.com> Signed-off-by:
-
Eric Dumazet authored
Brian reported crashes using IPv6 traffic with macvtap/veth combo. I tracked the crashes in neigh_hh_output() -> memcpy(skb->data - HH_DATA_MOD, hh->hh_data, HH_DATA_MOD); Neighbour code assumes headroom to push Ethernet header is at least 16 bytes. It appears macvtap has only 14 bytes available on arches where NET_IP_ALIGN is 0 (like x86) Effect is a corruption of 2 bytes right before skb->head, and possible crashes if accessing non existing memory. This fix should also increase IPv4 performance, as paranoid code in ip_finish_output2() wont have to call skb_realloc_headroom() Reported-by:
Brian Rak <brak@vultr.com> Tested-by:
-
David S. Miller authored
Merge tag 'mac80211-for-davem-2015-02-27' of git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211 Johannes Berg says: ==================== A few patches have accumulated, among them the fix for Linus's four-way-handshake problem. The others are various small fixes for problems all over, nothing really stands out. ==================== Signed-off-by:
-
- 28 Feb, 2015 5 commits
-
-
Eric W. Biederman authored
When applicable verify that the caller has permisson to the underlying network namespace for a newly created network device. Similary checks exist for the network namespace a network device will be created in. Fixes: 317f4810 ("rtnl: allow to create device with IFLA_LINK_NETNSID set") Signed-off-by:
"Eric W. Biederman" <ebiederm@xmission.com> Acked-by:
-
Eric W. Biederman authored
When applicable verify that the caller has permision to create a network device in another network namespace. This check is already present when moving a network device between network namespaces in setlink so all that is needed is to duplicate that check in newlink. This change almost backports cleanly, but there are context conflicts as the code that follows was added in v4.0-rc1 Fixes: b51642f6 net: Enable a userns root rtnl calls that are safe for unprivilged users Signed-off-by:
-
George McCollister authored
Prior to this patch, sending a packet with the source MAC address of one of the CPSW interfaces to one of the CPSW slave ports while it's configured in dual_emac mode would update the port_num field of the VLAN/Unicast Address Table Entry. This would cause it to discard all incoming traffic addressed to that MAC address, essentially rendering the port useless until the ALE table is cleared (by starting and stopping the interface or rebooting.) For example, if eth0 has a MAC address of 90:59:af:8f:43:e9 it will have an ALE table entry: 00 00 00 00 59 90 02 30 e9 43 8f af (VLAN Addr vlan_id=2 unicast type=0 port_num=0 addr=90:59:af:8f:43:e9) If you configure another device with the same MAC address and connect it to the first CPSW slave port and send some traffic the ALE table entry becomes: 04 00 00 00 59 90 02 30 e9 43 8f af (VLAN Addr vlan_id=2 unicast type=0 port_num=1 addr=90:59:af:8f:43:e9) >From this point forward all incoming traffic addressed to 90:59:af:8f:43:e9 will be dropped. Setting the SECURE bit for the VLAN/Unicast address table entry for each interface's MAC address corrects the problem. Signed-off-by:
George McCollister <george.mccollister@gmail.com> Signed-off-by:
-
Dan Carpenter authored
There is a discrepancy here because the niu_class_to_ethflow() returns zero on failure and one on success but the caller expected zero on success and negative on failure. The problem means that we allow the user to pass classes and flow_types which we don't want. I've looked at it a bit and I don't see it as a very serious bug. Signed-off-by:
Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by:
-
Arnd Bergmann authored
The smc91x driver traditionally gets configured at compile-time for whichever hardware it runs on. This no longer works on ARM as we continue to move to building all-in-one kernels. Most ARM configurations with this driver already use run-time configuration through DT or through platform_data, but a few have not been converted yet. I've checked all ARM boards that use this driver in their legacy board files, and converted the ones that were using compile-time configuration in smc91x.h to behave like the other ones and provide the interrupt polarity along with the MMIO configuration (width, stride) at platform device creation time. In particular, these combinations were previously selectable in Kconfig but in fact broken: - sa1100 assabet plus pleb - msm combined with any other armv6/v7 platform - pxa-idp combined with any non-DMA pxa variant - LogicPD PXA270 combined with any other pxa - nomadik combined with any other armv4/v5 platform, e.g. versatile. None of these seem critical enough to warrant a backport to stable, but it would be nice to clean this up for good. Signed-off-by:
Arnd Bergmann <arnd@arndb.de> ---- I would like the patch to get merged through netdev, after Robert and/or Linus have verified it on at least some hardware. There are a few other non-ARM platforms using this driver, I could do the same patch for those if we want to take it further. arch/arm/mach-msm/board-halibut.c | 8 ++++- arch/arm/mach-msm/board-qsd8x50.c | 8 ++++- arch/arm/mach-pxa/idp.c | 5 +++ arch/arm/mach-pxa/lpd270.c | 8 ++++- arch/arm/mach-realview/core.c | 7 ++++ arch/arm/mach-realview/realview_eb.c | 2 +- arch/arm/mach-sa1100/neponset.c | 6 ++++ arch/arm/mach-sa1100/pleb.c | 7 ++++ drivers/net/ethernet/smsc/smc91x.c | 9 +++-- drivers/net/ethernet/smsc/smc91x.h | 114 ++---------------------------------------------------------- 10 files changed, 57 insertions(+), 117 deletions(-) Tested-by:
Robert Jarzmik <robert.jarzmik@free.fr> Signed-off-by:
-
- 27 Feb, 2015 9 commits
-
-
Eric Dumazet authored
If a hash table has 128 slots and 16384 elems, expand to 256 slots takes more than one second. For larger sets, a soft lockup is detected. Holding cpu for that long, even in a work queue is a show stopper for non preemptable kernels. cond_resched() at strategic points to allow process scheduler to reschedule us. Signed-off-by:
Daniel Borkmann <daniel@iogearbox.net> Signed-off-by:
-
git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/netDavid S. Miller authored
Jeff Kirsher says: ==================== Intel Wired LAN Driver Updates 2015-02-26 This series contains fixes for i40e and i40evf only. Alexey Khoroshilov found a possible leak of 'cmd_buf' when copy_from_user() failed in i40e_dbg_command_write(), so resolved by calling kfree(). Shannon provides a fix to ensure the shift and bitwise precedences do not work backwards for us by adding parans. Fixed the driver by preventing the driver from allowing stray interrupts or causing system logs from un-handled interrupts by combining the ICR0 shutdown with the standard interrupt shutdown and add the interrupt clearing to the PCI shutdown path. Fixed an issue where a NVM write times out before a transaction can complete, so Shannon added logic to make another attempt by reacquiring the semaphore, then retry the write, if the one retry fails, we will then give up. Adds checks to pointers before their use to ensure we do not try to dereference NULL pointers when returning values from the AdminQ calls. Akeem adds a check to bail out if the device is already down when checking for Tx hang subtask. Anjali fixes TSO with more than 8 frags per segment issue. The hardware has some limitations which the driver needs to adhere to: 1) no more than 8 descriptors per packet on the wire 2) no header can span more than 3 descriptors If one of these events happens, the hardware will generate an internal error and freeze the Tx queue, so Anjali fixes this by linearizes the skb to avoid these situations. Fixed an issue where the per Traffic Class queue count was higher than queues enabled, which will fix a warning with multiple function mode where systems regularly have more cores than vectors. Fixed TCP/IPv6 over VXLAN Tx checksum offload, where we were checking the outer protocol flags and deciding the flow for the inner header. Jesse fixes a race condition in the transmit hang detection. Before we were having issues of false Tx hang detection, no the driver makes more direct with the checks for progress forward by directly checking the head write back address and tail register when determining progress. This avoids Tx hangs where the software gets behind, because we are directly checking hardware state when determining a hang state. Neerav fixes the transmit ring Qset handle when DCB reconfigures. The issue was when DCB is reconfigured to a single traffic class (TC) and the driver did not reset the Tx ring Qset handle to correct the mapping, which caused the Tx queue to disable timeouts. Also as part of DCB reconfiguration flow if the Tx queue disable times out, then issue a PF reset to do some level of recovery. Mitch stops flow director on shutdown because, in some cases, the hardware would continue to try to access the FDIR ring after entering D3Hot state, which would cause either PCIe errors or NMIs, depending upon the system configuration. * NOTE * I have verified that this series of patches for net will not cause any merge issues when you sync up your net tree with your net-next tree. ==================== Signed-off-by: -
Lendacky, Thomas authored
It is possible that the hardware may not have been properly shutdown before this driver gets control, through use by firmware, for example. Until the driver is loaded, interrupts associated with the hardware could go pending. When the IRQs are requested napi support has not been initialized yet, but the ISR will get control and schedule napi processing resulting in a kernel panic because the poll routine has not been set. Adjust the code so that the driver is fully ready to handle and process interrupts as soon as the IRQs are requested. This involves requesting and freeing IRQs during start and stop processing and ordering the napi add and delete calls appropriately. Also adjust the powerup and powerdown routines to match the start and stop routines in regards to the ordering of tasks, including napi related calls. Signed-off-by:
Tom Lendacky <thomas.lendacky@amd.com> Signed-off-by:
-
Luca Ceresoli authored
Just another AX88178-based 10/100/1000 USB-to-Ethernet dongle. This one shows up in lsusb as: "Sitecom Europe B.V. LN-028 Network USB 2.0 Adapter". Signed-off-by:
Luca Ceresoli <luca@lucaceresoli.net> Cc: Francois Romieu <romieu@fr.zoreil.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: linux-usb@vger.kernel.org Cc: linux-kernel@vger.kernel.org Signed-off-by:
-
David S. Miller authored
Daniel Borkmann says: ==================== rhashtable updates As discussed, I'm sending out rhashtable fixups for -net. I have a couple of more patches I was working on last week pending, i.e. to get rid of ht->nelems and ht->shift atomic operations which speed-up pure insertions/deletions, e.g. on my laptop I have 2 threads, inserting 7M entries each, that will reduce insertion time from ~1,450 ms to 865 ms (performance should even be better after removing the grow/shrink indirections). I guess that however is rather something for net-next. ==================== Signed-off-by: -
Daniel Borkmann authored
Currently, all real users of rhashtable default their grow and shrink decision functions to rht_grow_above_75() and rht_shrink_below_30(), so that there's currently no need to have this explicitly selectable. It can/should be generic and private inside rhashtable until a real use case pops up. Since we can make this private, we'll save us this additional indirection layer and can improve insertion/deletion time as well. Reference: http://patchwork.ozlabs.org/patch/443040/Suggested-by:
Thomas Graf <tgraf@suug.ch> Signed-off-by:
-
Daniel Borkmann authored
While commit c0c09bfd ("rhashtable: avoid unnecessary wakeup for worker queue") rightfully moved part of the decision making of whether we should expand or shrink from the expand/shrink functions themselves into insert/delete functions in order to avoid unnecessary worker wake-ups, it however introduced a regression by doing so. Before that change, if no max_shift was specified (= 0) on rhashtable initialization, rhashtable_expand() would just grow unconditionally and lets the available memory be the limiting factor. After that change, if no max_shift was specified, there would be _no_ expansion step at all. Given that netlink and tipc have a max_shift specified, it was not visible there, but Josh Hunt reported that if nft that starts out with a default element hint of 3 if not otherwise provided, would slow i.e. inserts down trememdously as it cannot grow larger to relax table occupancy. Given that the test case verifies shrinks/expands manually, we also must remove pointer to the helper functions to explicitly avoid parallel resizing on insertions/deletions. test_bucket_stats() and test_rht_lookup() could also be wrapped around rhashtable mutex to explicitly synchronize a walk from resizing, but I think that defeats the actual test case which intended to have explicit test steps, i.e. 1) inserts, 2) expands, 3) shrinks, 4) deletions, with object verification after each stage. Reported-by:
Josh Hunt <johunt@akamai.com> Fixes: c0c09bfd ("rhashtable: avoid unnecessary wakeup for worker queue") Signed-off-by:
-
Michael S. Tsirkin authored
The 2 that we use for copy_to_iter comes from sizeof(u16), it used to be that way before the iov iter update. Fix it up, making it obvious the size of stack access is right. Signed-off-by:
Michael S. Tsirkin <mst@redhat.com> Signed-off-by:
-
Michael S. Tsirkin authored
Recent iterator-related changes in vhost made it harder to follow the logic fixing up the header. In fact, the fixup always happens at the same offset: sizeof(virtio_net_hdr): sometimes the fixup iterator is updated by copy_to_iter, sometimes-by iov_iter_advance. Rearrange code to make this obvious. Signed-off-by:
-
