- 24 Jun, 2022 8 commits
-
-
Michael Walle authored
Instead of open-coding the bad characters replacement in the hwmon name, use the new hwmon_sanitize_name(). Signed-off-by:
Michael Walle <michael@walle.cc> Acked-by:
Russell King (Oracle) <rmk+kernel@armlinux.org.uk> Acked-by:
Guenter Roeck <linux@roeck-us.net> Signed-off-by:
Jakub Kicinski <kuba@kernel.org>
-
Jakub Kicinski authored
Jonathan Lemon says: ==================== Broadcom PTP PHY support This adds PTP support for the Broadcom PHY BCM54210E (and the specific variant BCM54213PE that the rpi-5.15 branch uses). This has only been tested on the RPI CM4, which has one port. There are other Broadcom chips which may benefit from using the same framework here, although with different register sets. ==================== Link: https://lore.kernel.org/r/20220622050454.878052-1-jonathan.lemon@gmail.comSigned-off-by:
-
Jonathan Lemon authored
The perout function is used to generate a 1PPS signal, synchronized to the PHC. This is accomplished by a using the hardware oneshot functionality, which is reset by a timer. The external timestamp function is set up for a 1PPS input pulse, and uses a timer to poll for temestamps. Both functions use the SYNC_OUT/SYNC_IN1 pin, so cannot run simultaneously. Co-developed-by:
Lasse Johnsen <l@ssejohnsen.me> Signed-off-by:
Jonathan Lemon <jonathan.lemon@gmail.com> Acked-by:
Richard Cochran <richardcochran@gmail.com> Signed-off-by:
-
Jonathan Lemon authored
This adds PTP support for BCM54210E Broadcom PHYs, in particular, the BCM54213PE, as used in the Rasperry PI CM4. It has only been tested on that hardware. Signed-off-by:
Florian Fainelli <f.fainelli@gmail.com> Signed-off-by:
-
Jonathan Lemon authored
Add 'struct bcm_ptp_private' to bcm54xx_phy_priv which points to an optional PTP structure attached to the PHY. This is allocated on probe if PHY PTP support is configured, and if the driver supports PTP for the specified PHY. Add the bcm_ptp_probe(), bcm_ptp_config_init() and bcn_ptp_stop() API functions to the bcm-phy library. Signed-off-by:
-
Jakub Kicinski authored
Russell King says: ==================== net: dsa: mv88e6xxx: get rid of SPEED_MAX This series does two things: 1. it gets rid of mv88e6065_port_set_speed_duplex() which is completely unused (do we support this device? I couldn't find it in the tables in chip.c) This has a max speed of 200Mbps which we don't support. 2. get rid of the SPEED_MAX constant, which is used to configure a DSA or CPU port to their maximum speed during initialisation. We no longer need this as we can derive the maximum port speed from the mac_capabilities instead. The reason for making this change is in preparation for phylink to be used by DSA for CPU ports. This omission has come back to bite us with the conversion of DSA drivers to phylink_pcs, since phylink_pcs won't get used unless phylink is being used. Particularly with this driver, it is very common for DT descriptions to omit the fixed-link details which means "use maximum speed". It will eventually be necessary to hoist the selection of "max speed" into the DSA layer (trivial) and also have a way for the DSA driver to tell the DSA layer which interface it should be using for these ports. ==================== Link: https://lore.kernel.org/r/YrGQBssOvQBZiDS4@shell.armlinux.org.ukSigned-off-by:
-
Russell King authored
Currently, all the device specific speed setting functions convert SPEED_MAX to the actual speed of the port. Rather than having each of the mv88e6xxx chip specifics handling SPEED_MAX, derive it from the mac_capabilities instead. This is only needed for CPU and DSA ports, so move the logic up into mv88e6xxx_setup_port() - which allows us to kill off all users of SPEED_MAX throughout the driver. Signed-off-by:
-
Russell King (Oracle) authored
Remove mv88e6065_port_set_speed_duplex() - this is never called, and thus is completely redundant. Signed-off-by:
-
- 23 Jun, 2022 22 commits
-
-
git://git.kernel.org/pub/scm/linux/kernel/git/netdev/netJakub Kicinski authored
No conflicts. Signed-off-by: -
git://git.kernel.org/pub/scm/linux/kernel/git/netdev/netLinus Torvalds authored
Pull networking fixes from Paolo Abeni: "Including fixes from bpf and netfilter. Current release - regressions: - netfilter: cttimeout: fix slab-out-of-bounds read in cttimeout_net_exit Current release - new code bugs: - bpf: ftrace: keep address offset in ftrace_lookup_symbols - bpf: force cookies array to follow symbols sorting Previous releases - regressions: - ipv4: ping: fix bind address validity check - tipc: fix use-after-free read in tipc_named_reinit - eth: veth: add updating of trans_start Previous releases - always broken: - sock: redo the psock vs ULP protection check - netfilter: nf_dup_netdev: fix skb_under_panic - bpf: fix request_sock leak in sk lookup helpers - eth: igb: fix a use-after-free issue in igb_clean_tx_ring - eth: ice: prohibit improper channel config for DCB - eth: at803x: fix null pointer dereference on AR9331 phy - eth: virtio_net: fix xdp_rxq_info bug after suspend/resume Misc: - eth: hinic: replace memcpy() with direct assignment" * tag 'net-5.19-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (47 commits) net: openvswitch: fix parsing of nw_proto for IPv6 fragments sock: redo the psock vs ULP protection check Revert "net/tls: fix tls_sk_proto_close executed repeatedly" virtio_net: fix xdp_rxq_info bug after suspend/resume igb: Make DMA faster when CPU is active on the PCIe link net: dsa: qca8k: reduce mgmt ethernet timeout net: dsa: qca8k: reset cpu port on MTU change MAINTAINERS: Add a maintainer for OCP Time Card hinic: Replace memcpy() with direct assignment Revert "drivers/net/ethernet/neterion/vxge: Fix a use-after-free bug in vxge-main.c" net: phy: smsc: Disable Energy Detect Power-Down in interrupt mode ice: ethtool: Prohibit improper channel config for DCB ice: ethtool: advertise 1000M speeds properly ice: Fix switchdev rules book keeping ice: ignore protocol field in GTP offload netfilter: nf_dup_netdev: add and use recursion counter netfilter: nf_dup_netdev: do not push mac header a second time selftests: netfilter: correct PKTGEN_SCRIPT_PATHS in nft_concat_range.sh net/tls: fix tls_sk_proto_close executed repeatedly erspan: do not assume transport header is always set ... -
git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmcLinus Torvalds authored
Pull MMC fixes from Ulf Hansson: - mtk-sd: Fix dma hang issues - sdhci-pci-o2micro: Fix card detect by dealing with debouncing * tag 'mmc-v5.19-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc: mmc: mediatek: wait dma stop bit reset to 0 mmc: sdhci-pci-o2micro: Fix card detect by dealing with debouncing
-
git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/soundLinus Torvalds authored
Pull sound fixes from Takashi Iwai: "All small changes, mostly device-specific: - A regression fix for PCM WC-page allocation on x86 - A regression fix for i915 audio component binding - Fixes for (longstanding) beep handling bug - Runtime PM fixes for Intel LPE HDMI audio - A couple of pending FireWire fixes - Usual HD-audio and USB-audio quirks, new Intel dspconf entries" * tag 'sound-5.19-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound: ALSA: hda/realtek: Add quirk for Clevo NS50PU ALSA: hda: Fix discovery of i915 graphics PCI device ALSA: hda/via: Fix missing beep setup ALSA: hda/conexant: Fix missing beep setup ALSA: memalloc: Drop x86-specific hack for WC allocations ALSA: hda/realtek: Add quirk for Clevo PD70PNT ALSA: x86: intel_hdmi_audio: use pm_runtime_resume_and_get() ALSA: x86: intel_hdmi_audio: enable pm_runtime and set autosuspend delay ALSA: hda: intel-nhlt: remove use of __func__ in dev_dbg ALSA: hda: intel-dspcfg: use SOF for UpExtreme and UpExtreme11 boards firewire: convert sysfs sprintf/snprintf family to sysfs_emit firewire: cdev: fix potential leak of kernel stack due to uninitialized value ALSA: hda/realtek: Apply fixup for Lenovo Yoga Duet 7 properly ALSA: hda/realtek - ALC897 headset MIC no sound ALSA: usb-audio: US16x08: Move overflow check before array access ALSA: hda/realtek: Add mute LED quirk for HP Omen laptop
-
Sixiang Chen authored
Add support for ethtool -p|--identify by enabling blinking of the panel LED if supported by the NIC firmware. Signed-off-by:
Sixiang Chen <sixiang.chen@corigine.com> Signed-off-by:
Simon Horman <simon.horman@corigine.com> Acked-by:
Paolo Abeni <pabeni@redhat.com>
-
Rosemarie O'Riorden authored
When a packet enters the OVS datapath and does not match any existing flows installed in the kernel flow cache, the packet will be sent to userspace to be parsed, and a new flow will be created. The kernel and OVS rely on each other to parse packet fields in the same way so that packets will be handled properly. As per the design document linked below, OVS expects all later IPv6 fragments to have nw_proto=44 in the flow key, so they can be correctly matched on OpenFlow rules. OpenFlow controllers create pipelines based on this design. This behavior was changed by the commit in the Fixes tag so that nw_proto equals the next_header field of the last extension header. However, there is no counterpart for this change in OVS userspace, meaning that this field is parsed differently between OVS and the kernel. This is a problem because OVS creates actions based on what is parsed in userspace, but the kernel-provided flow key is used as a match criteria, as described in Documentation/networking/openvswitch.rst. This leads to issues such as packets incorrectly matching on a flow and thus the wrong list of actions being applied to the packet. Such changes in packet parsing cannot be implemented without breaking the userspace. The offending commit is partially reverted to restore the expected behavior. The change technically made sense and there is a good reason that it was implemented, but it does not comply with the original design of OVS. If in the future someone wants to implement such a change, then it must be user-configurable and disabled by default to preserve backwards compatibility with existing OVS versions. Cc: stable@vger.kernel.org Fixes: fa642f08 ("openvswitch: Derive IP protocol number for IPv6 later frags") Link: https://docs.openvswitch.org/en/latest/topics/design/#fragmentsSigned-off-by:
Rosemarie O'Riorden <roriorden@redhat.com> Acked-by:
Eelco Chaudron <echaudro@redhat.com> Link: https://lore.kernel.org/r/20220621204845.9721-1-roriorden@redhat.comSigned-off-by:
-
Jakub Kicinski authored
Commit 8a59f9d1 ("sock: Introduce sk->sk_prot->psock_update_sk_prot()") has moved the inet_csk_has_ulp(sk) check from sk_psock_init() to the new tcp_bpf_update_proto() function. I'm guessing that this was done to allow creating psocks for non-inet sockets. Unfortunately the destruction path for psock includes the ULP unwind, so we need to fail the sk_psock_init() itself. Otherwise if ULP is already present we'll notice that later, and call tcp_update_ulp() with the sk_proto of the ULP itself, which will most likely result in the ULP looping its callbacks. Fixes: 8a59f9d1 ("sock: Introduce sk->sk_prot->psock_update_sk_prot()") Signed-off-by:
John Fastabend <john.fastabend@gmail.com> Reviewed-by:
Jakub Sitnicki <jakub@cloudflare.com> Tested-by:
-
Jakub Kicinski authored
This reverts commit 69135c57. This commit was just papering over the issue, ULP should not get ->update() called with its own sk_prot. Each ULP would need to add this check. Fixes: 69135c57 ("net/tls: fix tls_sk_proto_close executed repeatedly") Signed-off-by:
-
git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queueJakub Kicinski authored
Tony Nguyen says: ==================== 40GbE Intel Wired LAN Driver Updates 2022-06-21 This series contains updates to i40e driver only. Mateusz adds support for using the speed option in ethtool. Minghao Chi removes unneeded synchronize_irq() calls. Bernard Zhao removes unneeded NULL check. * '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/next-queue: intel/i40e: delete if NULL check before dev_kfree_skb i40e: Remove unnecessary synchronize_irq() before free_irq() i40e: Add support for ethtool -s <interface> speed <speed in Mb> ==================== Link: https://lore.kernel.org/r/20220621225930.632741-1-anthony.l.nguyen@intel.comSigned-off-by:
-
Stephan Gerhold authored
The following sequence currently causes a driver bug warning when using virtio_net: # ip link set eth0 up # echo mem > /sys/power/state (or e.g. # rtcwake -s 10 -m mem) <resume> # ip link set eth0 down Missing register, driver bug WARNING: CPU: 0 PID: 375 at net/core/xdp.c:138 xdp_rxq_info_unreg+0x58/0x60 Call trace: xdp_rxq_info_unreg+0x58/0x60 virtnet_close+0x58/0xac __dev_close_many+0xac/0x140 __dev_change_flags+0xd8/0x210 dev_change_flags+0x24/0x64 do_setlink+0x230/0xdd0 ... This happens because virtnet_freeze() frees the receive_queue completely (including struct xdp_rxq_info) but does not call xdp_rxq_info_unreg(). Similarly, virtnet_restore() sets up the receive_queue again but does not call xdp_rxq_info_reg(). Actually, parts of virtnet_freeze_down() and virtnet_restore_up() are almost identical to virtnet_close() and virtnet_open(): only the calls to xdp_rxq_info_(un)reg() are missing. This means that we can fix this easily and avoid such problems in the future by just calling virtnet_close()/open() from the freeze/restore handlers. Aside from adding the missing xdp_rxq_info calls the only difference is that the refill work is only cancelled if netif_running(). However, this should not make any functional difference since the refill work should only be active if the network interface is actually up. Fixes: 754b8a21 ("virtio_net: setup xdp_rxq_info") Signed-off-by:
Stephan Gerhold <stephan.gerhold@kernkonzept.com> Acked-by:
Jesper Dangaard Brouer <brouer@redhat.com> Acked-by:
Jason Wang <jasowang@redhat.com> Link: https://lore.kernel.org/r/20220621114845.3650258-1-stephan.gerhold@kernkonzept.comSigned-off-by:
-
git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queueJakub Kicinski authored
Tony Nguyen says: ==================== Intel Wired LAN Driver Updates 2022-06-21 This series contains updates to ice driver only. Marcin fixes GTP filters by allowing ignoring of the inner ethertype field. Wojciech adds VSI handle tracking in order to properly distinguish similar filters for removal. Anatolii removes ability to set 1000baseT and 1000baseX fields concurrently which caused link issues. He also disallows setting channels to less than the number of Traffic Classes which would cause NULL pointer dereference. * '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue: ice: ethtool: Prohibit improper channel config for DCB ice: ethtool: advertise 1000M speeds properly ice: Fix switchdev rules book keeping ice: ignore protocol field in GTP offload ==================== Link: https://lore.kernel.org/r/20220621224756.631765-1-anthony.l.nguyen@intel.comSigned-off-by:
-
Eric Dumazet authored
saddr and daddr are set but not used. Fixes: ba44f818 ("raw: use more conventional iterators") Reported-by:
kernel test robot <lkp@intel.com> Signed-off-by:
Eric Dumazet <edumazet@google.com> Acked-by:
-
Kai-Heng Feng authored
Intel I210 on some Intel Alder Lake platforms can only achieve ~750Mbps Tx speed via iperf. The RR2DCDELAY shows around 0x2xxx DMA delay, which will be significantly lower when 1) ASPM is disabled or 2) SoC package c-state stays above PC3. When the RR2DCDELAY is around 0x1xxx the Tx speed can reach to ~950Mbps. According to the I210 datasheet "8.26.1 PCIe Misc. Register - PCIEMISC", "DMA Idle Indication" doesn't seem to tie to DMA coalesce anymore, so set it to 1b for "DMA is considered idle when there is no Rx or Tx AND when there are no TLPs indicating that CPU is active detected on the PCIe link (such as the host executes CSR or Configuration register read or write operation)" and performing Tx should also fall under "active CPU on PCIe link" case. In addition to that, commit b6e0c419 ("igb: Move DMA Coalescing init code to separate function.") seems to wrongly changed from enabling E1000_PCIEMISC_LX_DECISION to disabling it, also fix that. Fixes: b6e0c419 ("igb: Move DMA Coalescing init code to separate function.") Signed-off-by:
Kai-Heng Feng <kai.heng.feng@canonical.com> Tested-by: Gurucharan <gurucharanx.g@intel.com> (A Contingent worker at Intel) Signed-off-by:
Tony Nguyen <anthony.l.nguyen@intel.com> Link: https://lore.kernel.org/r/20220621221056.604304-1-anthony.l.nguyen@intel.comSigned-off-by:
-
Bhadram Varka authored
Add support multi-gigabit and single-port Ethernet PHY transceiver (AQR113C). Signed-off-by:
Bhadram Varka <vbhadram@nvidia.com> Link: https://lore.kernel.org/r/20220621034027.56508-1-vbhadram@nvidia.comSigned-off-by:
-
Russell King (Oracle) authored
Use the recently introduced mdiodev accessors for the lynx PCS. Signed-off-by:
-
Christian Marangi authored
The current mgmt ethernet timeout is set to 100ms. This value is too big and would slow down any mdio command in case the mgmt ethernet packet have some problems on the receiving part. Reduce it to just 5ms to handle case when some operation are done on the master port that would cause the mgmt ethernet to not work temporarily. Fixes: 5950c7c0 ("net: dsa: qca8k: add support for mgmt read/write in Ethernet packet") Signed-off-by:
Christian Marangi <ansuelsmth@gmail.com> Link: https://lore.kernel.org/r/20220621151633.11741-1-ansuelsmth@gmail.comSigned-off-by:
-
Christian Marangi authored
It was discovered that the Documentation lacks of a fundamental detail on how to correctly change the MAX_FRAME_SIZE of the switch. In fact if the MAX_FRAME_SIZE is changed while the cpu port is on, the switch panics and cease to send any packet. This cause the mgmt ethernet system to not receive any packet (the slow fallback still works) and makes the device not reachable. To recover from this a switch reset is required. To correctly handle this, turn off the cpu ports before changing the MAX_FRAME_SIZE and turn on again after the value is applied. Fixes: f58d2598 ("net: dsa: qca8k: implement the port MTU callbacks") Cc: stable@vger.kernel.org Signed-off-by:
-
Jiang Jian authored
there is an unexpected word "the" in the comments that need to be dropped file: ./drivers/isdn/hardware/mISDN/hfcsusb.c line: 1560 /* set USB_SIZE_I to match the the wMaxPacketSize for ISO transfers */ changed to /* set USB_SIZE_I to match the wMaxPacketSize for ISO transfers */ Signed-off-by:
Jiang Jian <jiangjian@cdjrlc.com> Link: https://lore.kernel.org/r/20220621114529.108079-1-jiangjian@cdjrlc.comSigned-off-by:
-
Jiang Jian authored
there is an unexpected word "the" in the comments that need to be removed Signed-off-by:
Alex Elder <elder@linaro.org> Link: https://lore.kernel.org/r/20220621085001.61320-1-jiangjian@cdjrlc.comSigned-off-by:
-
Jiang Jian authored
there is an unexpected word "the" in the comments that need to be removed Signed-off-by:
-
Vadim Fedorenko authored
I've been contributing and reviewing patches for ptp_ocp driver for some time and I'm taking care of it's github mirror. On Jakub's suggestion, I would like to step forward and become a maintainer for this driver. This patch adds a dedicated entry to MAINTAINERS. Signed-off-by:
Vadim Fedorenko <vadfed@fb.com> Acked-by:
-
Yu Zhe authored
Remove unnecessary void* type castings. Signed-off-by:
Yu Zhe <yuzhe@nfschina.com> Acked-by:
Taehee Yoo <ap420073@gmail.com> Link: https://lore.kernel.org/r/20220621021648.2544-1-yuzhe@nfschina.comSigned-off-by:
-
- 22 Jun, 2022 10 commits
-
-
Linus Torvalds authored
Merge tag 'linux-kselftest-fixes-5.19-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/shuah/linux-kselftest Pull Kselftest fixes from Shuah Khan: "Compile time fixes and run-time resources leaks: - Fix clang cross compilation - Fix resource leak when return error - fix compile error for dma_map_benchmark - Fix regression - make use of GUP_TEST_FILE macro" * tag 'linux-kselftest-fixes-5.19-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/shuah/linux-kselftest: selftests: make use of GUP_TEST_FILE macro selftests: vm: Fix resource leak when return error selftests dma: fix compile error for dma_map_benchmark selftests: Fix clang cross compilation
-
Kees Cook authored
Under CONFIG_FORTIFY_SOURCE=y and CONFIG_UBSAN_BOUNDS=y, Clang is bugged here for calculating the size of the destination buffer (0x10 instead of 0x14). This copy is a fixed size (sizeof(struct fw_section_info_st)), with the source and dest being struct fw_section_info_st, so the memcpy should be safe, assuming the index is within bounds, which is UBSAN_BOUNDS's responsibility to figure out. Avoid the whole thing and just do a direct assignment. This results in no change to the executable code. [This is a duplicate of commit 2c0ab32b ("hinic: Replace memcpy() with direct assignment") which was applied to net-next.] Cc: Nick Desaulniers <ndesaulniers@google.com> Cc: Tom Rix <trix@redhat.com> Cc: llvm@lists.linux.dev Link: https://github.com/ClangBuiltLinux/linux/issues/1592Signed-off-by:
Kees Cook <keescook@chromium.org> Reviewed-by:
Gustavo A. R. Silva <gustavoars@kernel.org> Tested-by: Nathan Chancellor <nathan@kernel.org> # build Link: https://lore.kernel.org/r/20220616052312.292861-1-keescook@chromium.orgSigned-off-by:
-
Tim Crawford authored
Fixes headset detection on Clevo NS50PU. Signed-off-by:
Tim Crawford <tcrawford@system76.com> Cc: <stable@vger.kernel.org> Link: https://lore.kernel.org/r/20220622150017.9897-1-tcrawford@system76.comSigned-off-by:
Takashi Iwai <tiwai@suse.de>
-
https://github.com/martinetd/linuxLinus Torvalds authored
Pull 9pfs fixes from Dominique Martinet: "A couple of fid refcount and fscache fixes: - fid refcounting was incorrect in some corner cases and would leak resources, only freed at umount time. The first three commits fix three such cases - 'cache=loose' or fscache was broken when trying to write a partial page to a file with no read permission since the rework a few releases ago. The fix taken here is just to restore old behavior of using the special 'writeback_fid' for such reads, which is open as root/RDWR and such not get complains that we try to read on a WRONLY fid. Long-term it'd be nice to get rid of this and not issue the read at all (skip cache?) in such cases, but that direction hasn't progressed" * tag '9p-for-5.19-rc4' of https://github.com/martinetd/linux: 9p: fix EBADF errors in cached mode 9p: Fix refcounting during full path walks for fid lookups 9p: fix fid refcount leak in v9fs_vfs_get_link 9p: fix fid refcount leak in v9fs_vfs_atomic_open_dotl -
Jakub Kicinski authored
This reverts commit 8fc74d18. BAR0 is the main (only?) register bank for this device. We most obviously can't unmap it before the netdev is unregistered. This was pointed out in review but the patch got reposted and merged, anyway. The author of the patch was only testing it with a QEMU model, which I presume does not emulate enough for the netdev to be brought up (author's replies are not visible in lore because they kept sending their emails in HTML). Link: https://lore.kernel.org/all/20220616085059.680dc215@kernel.org/ Fixes: 8fc74d18 ("drivers/net/ethernet/neterion/vxge: Fix a use-after-free bug in vxge-main.c") Signed-off-by:
David S. Miller <davem@davemloft.net>
-
David S. Miller authored
Kuniyuki Iwashima says: ==================== af_unix: Introduce per-netns socket hash table. This series replaces unix_socket_table with a per-netns hash table and reduces lock contention and time on iterating over the list. Note the 3rd-6th patches can be a single patch, but for ease of review, they are split into small changes without breakage. Changes: v3: 6th: * Remove unix_table_locks from comments. * Remove missed spin_unlock(&unix_table_locks) in unix_lookup_by_ino() (kernel test robot) v2: https://lore.kernel.org/netdev/20220620185151.65294-1-kuniyu@amazon.com/ 3rd: * Update changelog * Remove holes from per-netns hash table structure * Use kvmalloc_array() instead of kmalloc() (Eric Dumazet) * Remove unnecessary parts in af_unix_init() (Eric Dumazet) * Move `err_sysctl` label into ifdef block (kernel test robot) * Remove struct netns_unix from struct net if CONFIG_UNIX is disabled 4th: * Use spin_lock_nested() (kernel test robot) v1: https://lore.kernel.org/netdev/20220616234714.4291-1-kuniyu@amazon.com/ ==================== Signed-off-by: -
Kuniyuki Iwashima authored
unix_table_locks are to protect the global hash table, unix_socket_table. The previous commit removed it, so let's clean up the unnecessary locks. Here is a test result on EC2 c5.9xlarge where 10 processes run concurrently in different netns and bind 100,000 sockets for each. without this series : 1m 38s with this series : 11s It is ~10x faster because the global hash table is split into 10 netns in this case. Signed-off-by:
Kuniyuki Iwashima <kuniyu@amazon.com> Signed-off-by:
-
Kuniyuki Iwashima authored
This commit replaces the global hash table with a per-netns one and removes the global one. We now link a socket in each netns's hash table so we can save some netns comparisons when iterating through a hash bucket. Signed-off-by:
-
Kuniyuki Iwashima authored
This commit adds extra spin_lock/spin_unlock() for a per-netns hash table inside the existing ones for unix_table_locks. As of this commit, sockets are still linked in the global hash table. After putting sockets in a per-netns hash table and removing the old one in the next patch, we remove the global locks in the last patch. Signed-off-by:
-
Kuniyuki Iwashima authored
This commit adds a per netns hash table for AF_UNIX, which size is fixed as UNIX_HASH_SIZE for now. The first implementation defines a per-netns hash table as a single array of lock and list: struct unix_hashbucket { spinlock_t lock; struct hlist_head head; }; struct netns_unix { struct unix_hashbucket *hash; ... }; But, Eric pointed out memory cost that the structure has holes because of sizeof(spinlock_t), which is 4 (or more if LOCKDEP is enabled). [0] It could be expensive on a host with thousands of netns and few AF_UNIX sockets. For this reason, a per-netns hash table uses two dense arrays. struct unix_table { spinlock_t *locks; struct hlist_head *buckets; }; struct netns_unix { struct unix_table table; ... }; Note the length of the list has a significant impact rather than lock contention, so having shared locks can be an option. But, per-netns locks and lists still perform better than the global locks and per-netns lists. [1] Also, this patch adds a change so that struct netns_unix disappears from struct net if CONFIG_UNIX is disabled. [0]: https://lore.kernel.org/netdev/CANn89iLVxO5aqx16azNU7p7Z-nz5NrnM5QTqOzueVxEnkVTxyg@mail.gmail.com/ [1]: https://lore.kernel.org/netdev/20220617175215.1769-1-kuniyu@amazon.com/Signed-off-by:
-
