- 12 Feb, 2024 39 commits
-
-
Eric Dumazet authored
nf_conntrack_cleanup_net_list() is calling synchronize_net() while RTNL is not held. This effectively calls synchronize_rcu(). synchronize_rcu() is much slower than synchronize_rcu_expedited(), and cleanup_net() is currently single threaded. In many workloads we want cleanup_net() to be faster, in order to free memory and various sysfs and procfs entries as fast as possible. Signed-off-by:
Eric Dumazet <edumazet@google.com> Cc: Pablo Neira Ayuso <pablo@netfilter.org> Cc: Jozsef Kadlecsik <kadlec@netfilter.org> Cc: Florian Westphal <fw@strlen.de> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Eric Dumazet authored
cleanup_net() is calling synchronize_rcu() right before acquiring RTNL. synchronize_rcu() is much slower than synchronize_rcu_expedited(), and cleanup_net() is currently single threaded. In many workloads we want cleanup_net() to be fast, in order to free memory and various sysfs and procfs entries as fast as possible. Signed-off-by:
-
Eric Dumazet authored
tnode_free() should use synchronize_net() instead of syncronize_rcu() to release RTNL sooner. Signed-off-by:
David Ahern <dsahern@kernel.org> Signed-off-by:
-
Eric Dumazet authored
br_vlan_flush() and nbp_vlan_flush() should use synchronize_net() instead of syncronize_rcu() to release RTNL sooner. Signed-off-by:
Nikolay Aleksandrov <razor@blackwall.org> Signed-off-by:
-
Eric Dumazet authored
dev_change_name() holds RTNL, we better use synchronize_net() instead of plain synchronize_rcu(). Signed-off-by:
-
Eric Dumazet authored
As discussed in the past (commit 2d3916f3 ("ipv6: fix skb drops in igmp6_event_query() and igmp6_event_report()")) I think the synchronize_net() call in ipv6_mc_down() is not needed. Under load, synchronize_net() can last between 200 usec and 5 ms. KASAN seems to agree as well. Fixes: f185de28 ("mld: add new workqueues for process mld events") Signed-off-by:
-
David S. Miller authored
Suraj Jaiswal says: ==================== Ethernet common fault IRQ support Changes since v13: - Update correct sender email Changes since v12: - Update correct sender email Changes since v11: - Update debug message print Changes since v10: - Update commit message Changes since v9: - prevent race condition of safety IRQ handling Changes since v8: - Use shared IRQ for sfty - update error message Changes since v7: - Add support of common sfty irq on stmmac_request_irq_multi_msi. - Remove uncecessary blank line. Changes since v6: - use name sfty_irq instead of safety_common_irq. Changes since v5: - Add description of ECC, DPP, FSM Changes since v4: - Fix DT_CHECKER warning - use name safety for the IRQ. ==================== Signed-off-by: -
Suraj Jaiswal authored
Add support to listen HW safety IRQ like ECC(error correction code), DPP(data path parity), FSM(finite state machine) fault in common IRQ line. Signed-off-by:
Suraj Jaiswal <quic_jsuraj@quicinc.com> Reviewed-by:
Serge Semin <fancer.lancer@gmail.com> Signed-off-by:
-
Suraj Jaiswal authored
Add binding doc for safety IRQ. The safety IRQ will be triggered for ECC(error correction code), DPP(data path parity), FSM(finite state machine) error. Signed-off-by:
Rob Herring <robh@kernel.org> Signed-off-by:
-
Heiner Kallweit authored
The reported bug is caused by using mii_eee_cap1_mod_linkmode_t() with an uninitialized bitmap. Fix this by zero-initializing the struct containing the bitmap. Fixes: 9bc79134 ("tg3: convert EEE handling to use linkmode bitmaps") Reported-by:
Srikanth Aithal <sraithal@amd.com> Tested-by:
Heiner Kallweit <hkallweit1@gmail.com> Reviewed-by:
Pavan Chebbi <pavan.chebbi@broadcom.com> Reviewed-by:
Andrew Lunn <andrew@lunn.ch> Signed-off-by:
-
David S. Miller authored
Luiz Angelo Daros de Luca says: ==================== net: dsa: realtek: variants to drivers, interfaces to a common module The current driver consists of two interface modules (SMI and MDIO) and two family/variant modules (RTL8365MB and RTL8366RB). The SMI and MDIO modules serve as the platform and MDIO drivers, respectively, calling functions from the variant modules. In this setup, one interface module can be loaded independently of the other, but both variants must be loaded (if not disabled at build time) for any type of interface. This approach doesn't scale well, especially with the addition of more switch variants (e.g., RTL8366B), leading to loaded but unused modules. Additionally, this also seems upside down, as the specific driver code normally depends on the more generic functions and not the other way around. Each variant module was converted into real drivers, serving as both a platform driver (for switches connected using the SMI interface) and an MDIO driver (for MDIO-connected switches). The relationship between the variant and interface modules is reversed, with the variant module now calling both interface functions (if not disabled at build time). While in most devices only one interface is likely used, the interface code is significantly smaller than a variant module, consuming fewer resources than the previous code. With variant modules now functioning as real drivers, compatible strings are published only in a single variant module, preventing conflicts. The patch series introduces a new common module for functions shared by both variants. This module also absorbs the two previous interface modules, as they would always be loaded anyway. The series relocates the user MII driver from realtek-smi to rtl83xx. It is now used by MDIO-connected switches instead of the generic DSA driver. There's a change in how this driver locates the MDIO node. It now only searches for a child node named "mdio". The dsa_switch in realtek_priv->ds is now embedded in the struct. It is always in use and avoids dynamic memory allocation. Testing has been performed with an RTL8367S (rtl8365mb) using MDIO interface and an RTL8366RB (rtl8366) with SMI interface. ==================== Signed-off-by: -
Luiz Angelo Daros de Luca authored
Embed dsa_switch within realtek_priv to eliminate the need for a second memory allocation. Suggested-by:
Alvin Šipraga <alsi@bang-olufsen.dk> Signed-off-by:
Luiz Angelo Daros de Luca <luizluca@gmail.com> Reviewed-by:
Vladimir Oltean <olteanv@gmail.com> Reviewed-by:
Linus Walleij <linus.walleij@linaro.org> Signed-off-by:
-
Luiz Angelo Daros de Luca authored
The realtek-mdio will now use this driver instead of the generic DSA driver ("dsa user smi"), which should not be used with OF[1]. With a single ds_ops for both interfaces, the ds_ops in realtek_priv is no longer necessary. Now, the realtek_variant.ds_ops can be used directly. The realtek_priv.setup_interface() has been removed as we can directly call the new common function. [1] https://lkml.kernel.org/netdev/20220630200423.tieprdu5fpabflj7@bang-olufsen.dk/T/Signed-off-by: -
Luiz Angelo Daros de Luca authored
In the user MDIO driver, despite numerous references to SMI, including its compatible string, there's nothing inherently specific about the SMI interface in the user MDIO bus. Consequently, the code has been migrated to the rtl83xx module. All references to SMI have been eliminated. The MDIO bus id was changed from Realtek-<switch id> to the switch devname suffixed with :user_mii, giving more information about the bus it is referencing. Signed-off-by:
-
Luiz Angelo Daros de Luca authored
Remove the line assigning dev.of_node in mdio_bus as subsequent of_mdiobus_register will always overwrite it. As discussed in [1], allow the DSA core to be simplified, by not assigning ds->user_mii_bus when the MDIO bus is described in OF, as it is unnecessary. Since commit 3b73a7b8 ("net: mdio_bus: add refcounting for fwnodes to mdiobus"), we can put the "mdio" node just after the MDIO bus registration. [1] https://lkml.kernel.org/netdev/20231213120656.x46fyad6ls7sqyzv@skbuf/T/#uSigned-off-by:
-
Luiz Angelo Daros de Luca authored
The binding docs requires for SMI-connected devices that the switch must have a child node named "mdio" and with a compatible string of "realtek,smi-mdio". Meanwile, for MDIO-connected switches, the binding docs only requires a child node named "mdio". This patch changes the driver to use the common denominator for both interfaces, looking for the MDIO node by name, ignoring the compatible string. Signed-off-by:
Florian Fainelli <florian.fainelli@broadcom.com> Reviewed-by:
-
Luiz Angelo Daros de Luca authored
Since rtl83xx and realtek-{smi,mdio} are always loaded together, we can optimize resource usage by consolidating them into a single module. Signed-off-by: -
Luiz Angelo Daros de Luca authored
Some code can be shared between both interface modules (MDIO and SMI) and among variants. These interface functions migrated to a common module: - rtl83xx_lock - rtl83xx_unlock - rtl83xx_probe - rtl83xx_register_switch - rtl83xx_unregister_switch - rtl83xx_shutdown - rtl83xx_remove The reset during probe was moved to the end of the common probe. This way, we avoid a reset if anything else fails. Signed-off-by:
-
Luiz Angelo Daros de Luca authored
Instead of copying values from the variant, we can keep a reference in realtek_priv. This is a preliminary change for sharing code betwen interfaces. It will allow to move most of the probe into a common module while still allow code specific to each interface to read variant fields. Signed-off-by:
-
Luiz Angelo Daros de Luca authored
Previously, the interface modules realtek-smi and realtek-mdio served as a platform and an MDIO driver, respectively. Each interface module redundantly specified the same compatible strings for both variants and referenced symbols from the variants. Now, each variant module has been transformed into a unified driver serving both as a platform and an MDIO driver. This modification reverses the relationship between the interface and variant modules, with the variant module now utilizing symbols from the interface modules. Signed-off-by:
-
Luiz Angelo Daros de Luca authored
Create a namespace to group the exported symbols. Signed-off-by:
-
Luiz Angelo Daros de Luca authored
It was never used and never referenced. Signed-off-by:
-
Abhishek Chauhan authored
TSO and TBS cannot co-exist. TBS requires special descriptor to be allocated at bootup. Initialising Tx queues at probe to support TSO and TBS can help in allocating those resources at bootup. TX queues with TBS can support etf qdisc hw offload. This is similar to the patch raised by NXP commit 3b12ec8f ("net: stmmac: dwmac-imx: set TSO/TBS TX queues default settings") Tested-by: Andrew Halaney <ahalaney@redhat.com> # sa8775p-ride Signed-off-by:
Abhishek Chauhan <quic_abchauha@quicinc.com> Reviewed-by:
Jeff Johnson <quic_jjohnson@quicinc.com> Reviewed-by:
Andrew Halaney <ahalaney@redhat.com> Signed-off-by:
-
David S. Miller authored
Kui-Feng Lee says: ==================== Remove expired routes with a separated list of routes. This patchset is resent due to previous reverting. [1] FIB6 GC walks trees of fib6_tables to remove expired routes. Walking a tree can be expensive if the number of routes in a table is big, even if most of them are permanent. Checking routes in a separated list of routes having expiration will avoid this potential issue. Background ========== The size of a Linux IPv6 routing table can become a big problem if not managed appropriately. Now, Linux has a garbage collector to remove expired routes periodically. However, this may lead to a situation in which the routing path is blocked for a long period due to an excessive number of routes. For example, years ago, there is a commit c7bb4b89 ("ipv6: tcp: drop silly ICMPv6 packet too big messages"). The root cause is that malicious ICMPv6 packets were sent back for every small packet sent to them. These packets add routes with an expiration time that prompts the GC to periodically check all routes in the tables, including permanent ones. Why Route Expires ================= Users can add IPv6 routes with an expiration time manually. However, the Neighbor Discovery protocol may also generate routes that can expire. For example, Router Advertisement (RA) messages may create a default route with an expiration time. [RFC 4861] For IPv4, it is not possible to set an expiration time for a route, and there is no RA, so there is no need to worry about such issues. Create Routes with Expires ========================== You can create routes with expires with the command. For example, ip -6 route add 2001:b000:591::3 via fe80::5054:ff:fe12:3457 \ dev enp0s3 expires 30 The route that has been generated will be deleted automatically in 30 seconds. GC of FIB6 ========== The function called fib6_run_gc() is responsible for performing garbage collection (GC) for the Linux IPv6 stack. It checks for the expiration of every route by traversing the trees of routing tables. The time taken to traverse a routing table increases with its size. Holding the routing table lock during traversal is particularly undesirable. Therefore, it is preferable to keep the lock for the shortest possible duration. Solution ======== The cause of the issue is keeping the routing table locked during the traversal of large trees. To solve this problem, we can create a separate list of routes that have expiration. This will prevent GC from checking permanent routes. Result ====== We conducted a test to measure the execution times of fib6_gc_timer_cb() and observed that it enhances the GC of FIB6. During the test, we added permanent routes with the following numbers: 1000, 3000, 6000, and 9000. Additionally, we added a route with an expiration time. Here are the average execution times for the kernel without the patch. - 120020 ns with 1000 permanent routes - 308920 ns with 3000 ... - 581470 ns with 6000 ... - 855310 ns with 9000 ... The kernel with the patch consistently takes around 14000 ns to execute, regardless of the number of permanent routes that are installed. Majro changes from v5: - Force syncrhonize GC before query expired routes with "sysctl -wq net.ipv6.route.flush=1". Major changes from v4: - Fix the comment of fib6_add_gc_list(). Major changes from v3: - Move the checks of f6i->fib6_node to fib6_add_gc_list(). - Make spin_lock_bh() and spin_unlock_bh() stands out. - Explain the reason of the changes in the commit message of the patch 4. Major changes from v2: - Refactory the boilerplate checks in the test case. - check_rt_num() and check_rt_num_clean() Major changes from v1: - Reduce the numbers of routes (5) in the test cases to work with slow environments. Due to the failure on patchwork. - Remove systemd related commands in the test case. Major changes from the previous patchset [2]: - Split helpers. - fib6_set_expires() -> fib6_set_expires() and fib6_add_gc_list(). - fib6_clean_expires() -> fib6_clean_expires() and fib6_remove_gc_list(). - Fix rt6_add_dflt_router() to avoid racing of setting expires. - Remove unnecessary calling to fib6_clean_expires() in ip6_route_info_create(). - Add test cases of toggling routes between permanent and temporary and handling routes from RA messages. - Clean up routes by deleting the existing device and adding a new one. - Fix a potential issue in modify_prefix_route(). ==================== Signed-off-by:
-
Kui-Feng Lee authored
Add tests of changing permanent routes to temporary routes and the reversed case to make sure GC working correctly in these cases. Add tests for the temporary routes from RA. The existing device will be deleted between tests to remove all routes associated with it, so that the earlier tests don't mess up the later ones. Reviewed-by:
Hangbin Liu <liuhangbin@gmail.com> Reviewed-by:
Kui-Feng Lee <thinker.li@gmail.com> Signed-off-by:
-
Kui-Feng Lee authored
Make the decision to set or clean the expires of a route based on the RTF_EXPIRES flag, rather than the value of the "expires" argument. This patch doesn't make difference logically, but make inet6_addr_modify() and modify_prefix_route() consistent. The function inet6_addr_modify() is the only caller of modify_prefix_route(), and it passes the RTF_EXPIRES flag and an expiration value. The RTF_EXPIRES flag is turned on or off based on the value of valid_lft. The RTF_EXPIRES flag is turned on if valid_lft is a finite value (not infinite, not 0xffffffff). Even if valid_lft is 0, the RTF_EXPIRES flag remains on. The expiration value being passed is equal to the valid_lft value if the flag is on. However, if the valid_lft value is infinite, the expiration value becomes 0 and the RTF_EXPIRES flag is turned off. Despite this, modify_prefix_route() decides to set the expiration value if the received expiration value is not zero. This mixing of infinite and zero cases creates an inconsistency. Reviewed-by:
-
Kui-Feng Lee authored
FIB6 GC walks trees of fib6_tables to remove expired routes. Walking a tree can be expensive if the number of routes in a table is big, even if most of them are permanent. Checking routes in a separated list of routes having expiration will avoid this potential issue. Reviewed-by:
-
Kui-Feng Lee authored
The route here is newly created. It is unnecessary to call fib6_clean_expires() on it. Suggested-by:
-
Kui-Feng Lee authored
Pass the duration of a lifetime (in seconds) to the function rt6_add_dflt_router() so that it can properly set the expiration time. The function ndisc_router_discovery() is the only one that calls rt6_add_dflt_router(), and it will later set the expiration time for the route created by rt6_add_dflt_router(). However, there is a gap of time between calling rt6_add_dflt_router() and setting the expiration time in ndisc_router_discovery(). During this period, there is a possibility that a new route may be removed from the routing table. By setting the correct expiration time in rt6_add_dflt_router(), we can prevent this from happening. The reason for setting RTF_EXPIRES in rt6_add_dflt_router() is to start the Garbage Collection (GC) timer, as it only activates when a route with RTF_EXPIRES is added to a table. Suggested-by:
-
Paolo Abeni authored
This test is time sensitive. It may fail on virtual machines and for debug builds. Similar to commit c41dfb0d ("selftests/net: ignore timing errors in so_txtime if KSFT_MACHINE_SLOW"), optionally suppress failure for timing errors (only). Signed-off-by:
Paolo Abeni <pabeni@redhat.com> Reviewed-by:
Willem de Bruijn <willemb@google.com> Signed-off-by:
-
David S. Miller authored
Shinas Rasheed says: ==================== add octeon_ep_vf driver This driver implements networking functionality of Marvell's Octeon PCI Endpoint NIC VF. This driver support following devices: * Network controller: Cavium, Inc. Device b203 * Network controller: Cavium, Inc. Device b403 * Network controller: Cavium, Inc. Device b103 * Network controller: Cavium, Inc. Device b903 * Network controller: Cavium, Inc. Device ba03 * Network controller: Cavium, Inc. Device bc03 * Network controller: Cavium, Inc. Device bd03 Changes: V7: - Separated octep_vf_get_if_stats from octep_vf_main.h to later patch in [1/8] - Moved introducing ndo_ops from [3/8] to [5/8] V6: https://lore.kernel.org/all/20240207065207.3092004-1-srasheed@marvell.com/ - Removed reuse of netif_tx_stop_all_queues, called implicitly in netif_tx_disable, when stopping netdev - Corrected error jump labels to have proper action-specific names in probe function - Removed singlethreaded workqueue implementation, since only tx timeout task is run. Run the same in the system workqueue - netdev_hold when tx_timeout happens to protect against free_netdev if race occurs between rmmod and a tx timeout. netdev_put the reference when timeout task ends to progress freeing netdev V5: https://lore.kernel.org/all/20240129050254.3047778-1-srasheed@marvell.com/ - Changed unchecked return types to void and removed unnecessary initializations in [2/8] patch. V4: https://lore.kernel.org/all/20240108124213.2966536-1-srasheed@marvell.com/ - Moved some stats from ethtool and added more to ndo_get_stats64 - Replaced code in IQ full check function to use helper from net/netdev_queues.h - Refactored code so that NETDEV_TX_BUSY is avoided V3: https://lore.kernel.org/all/20240105203823.2953604-1-srasheed@marvell.com/ - Removed UINT64_MAX, which is unused - Replaced masks and ULL declarations with GENMASK_ULL(), ULL() and other linux/bits.h macros, corrected declarations to conform to xmas tree format in patch [2/8] - Moved vfree and vzalloc null pointer casting corrections to patch [3/8], and corrected return values to follow standard kernel error codes in same - Set static budget of 64 for tx completion processing in NAPI - Replaces napi_complete and build_skb APIs to napi_complete_done and napi_build_skb APIs respectively - Replaced code with helper from net/netdev_queues.h to wake queues in TX completion processing - Removed duplicate reporting of TX/RX packets/bytes, which is already done during ndo_get_stats64 V2: https://lore.kernel.org/all/20231223134000.2906144-1-srasheed@marvell.com/ - Removed linux/version.h header file from inclusion in octep_vf_main.c - Corrected Makefile entry to include building octep_vf_mbox.c in [6/8] patch. - Removed redundant vzalloc pointer cast and vfree pointer check in [6/8] patch. V1: https://lore.kernel.org/all/20231221092844.2885872-1-srasheed@marvell.com/ ==================== Signed-off-by: -
Shinas Rasheed authored
add MAINTAINERS for octeon_ep_vf driver. Signed-off-by:
Shinas Rasheed <srasheed@marvell.com> Signed-off-by:
-
Shinas Rasheed authored
Add support for the following ethtool commands: ethtool -i|--driver devname ethtool devname ethtool -S|--statistics devname Signed-off-by:
-
Shinas Rasheed authored
Add support to enable MSI-x and register interrupts. Add support to process Tx and Rx traffic. Includes processing Tx completions and Rx refill. Signed-off-by:
-
Shinas Rasheed authored
Add support for ndo ops to set MAC address, change MTU, get stats. Add control path support to set MAC address, change MTU, get stats, set speed, get and set link mode. Signed-off-by:
-
Shinas Rasheed authored
Implement Tx/Rx ring resource allocation and cleanup. Signed-off-by:
-
Shinas Rasheed authored
Implement VF-PF mailbox to send all control commands from VF to PF and receive responses and notifications from PF to VF. Signed-off-by:
-
Shinas Rasheed authored
Implement hardware resource init and shutdown helper APIs, like hardware Tx/Rx queue init/enable/disable/reset. Signed-off-by:
-
Shinas Rasheed authored
Add driver framework and device setup and initialization for Octeon PCI Endpoint NIC VF. Add implementation to load module, initialize, register network device, cleanup and unload module. Signed-off-by:
-
- 10 Feb, 2024 1 commit
-
-
David S. Miller authored
Christian Marangi says: ==================== net: phy: Introduce PHY Package concept Idea of this big series is to introduce the concept of PHY package in DT and give PHY drivers a way to derive the base address from DT. The concept of PHY package is nothing new and is already a thing in the kernel with the API phy_package_join/leave/read/write. What is currently lacking is describing this in DT and better reference a base address to calculate offset from. In the scenario of a PHY package where multiple address are used and there isn't a way to get the base address of the PHY package from some regs, getting the information from DT is the only way. A possible example to this problem is this: ethernet-phy-package@0 { compatible = "qcom,qca8075-package"; #address-cells = <1>; #size-cells = <0>; reg = <0>; qcom,package-mode = "qsgmii"; ethernet-phy@1 { reg = <1>; }; phy4: ethernet-phy@4 { reg = <4>; }; }; The mdio parse functions are changed to address for this additional special node, the function is changed to simply detect this node and search also in this. (we match the node name to be "ethernet-phy-package") PHY driver can then use introduced helper of_phy_package_join to join the PHY to the PHY package and derive the base address from DT. Changes v7: - Rebase on top of net-next - Add Reviewed-by tag for DT patch - Change tx-driver-strength to tx-drive-strength - Drop driver reference in DT Changes v6: - Back to absolute PHY implementation - Correctly drop refcount for node on error condition and on PHY leave - Drop DT include patch in favor for 3 boolean vendor property - Fix Documentation problem for compatible and missing type and description - Drop redundand gpio-controller dependency and description - Skip scanphy with invalid PHY Package node and make reg mandatory - Rework fiber read status to use more generic function - Split qca808x LED generalization patch to permit easier review - Correctly return -EINVAL with wrong data passed to vendor property - Drop removing LED ops for qca807x PHY driver with gpio-controller Changes v5: - Rebase on top of net-next - Change implementation to base addr + offset in subnode - Adapt to all the changes and cleanup done to at803x Changes v4: - Rework DT implementation - Drop of autojoin support and rework to simple helper - Rework PHY driver to the new implementation - Add compatible for qca807x package - Further cleanup patches Changes v3: - Add back compatible implementation - Detach patch that can be handled separately (phy_package_mmd, phy_package extended) - Rework code to new simplified implementation with base addr + offset - Improve documentation with additional info and description Changes v2: - Drop compatible "ethernet-phy-package", use node name prefix matching instead - Improve DT example - Add reg for ethernet-phy-package - Drop phy-mode for ethernet-phy-package - Drop patch for generalization of phy-mode - Drop global-phy property (handle internally to the PHY driver) - Rework OF phy package code and PHY driver to handle base address - Fix missing of_node_put - Add some missing docs for added variables in struct - Move some define from dt-bindings include to PHY driver - Handle qsgmii validation in PHY driver - Fix wrong include for gpiolib - Drop reduntant version.h include ==================== Signed-off-by:
-
