- 25 Jan, 2023 12 commits
-
-
Stefan Raspl authored
We separate the code implementing the struct smcd_ops API in the ISM device driver from the functions that may be used by other exploiters of ISM devices. Note: We start out small, and don't offer the whole breadth of the ISM device for public use, as many functions are specific to or likely only ever used in the context of SMC-D. This is the third part of a bigger overhaul of the interfaces between SMC and ISM. Signed-off-by:
Stefan Raspl <raspl@linux.ibm.com> Signed-off-by:
Jan Karcher <jaka@linux.ibm.com> Signed-off-by:
Wenjia Zhang <wenjia@linux.ibm.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Stefan Raspl authored
Register the smc module with the new ism device driver API. This is the second part of a bigger overhaul of the interfaces between SMC and ISM. Signed-off-by:
-
Stefan Raspl authored
Add a new API that allows other drivers to concurrently access ISM devices. To do so, we introduce a new API that allows other modules to register for ISM device usage. Furthermore, we move the GID to struct ism, where it belongs conceptually, and rename and relocate struct smcd_event to struct ism_event. This is the first part of a bigger overhaul of the interfaces between SMC and ISM. Signed-off-by:
-
Stefan Raspl authored
Conceptually, a DMB is a structure that belongs to ISM devices. However, SMC currently 'owns' this structure. So future exploiters of ISM devices would be forced to include SMC headers to work - which is just weird. Therefore, we switch ISM to struct ism_dmb, introduce a new public header with the definition (will be populated with further API calls later on), and, add a thin wrapper to please SMC. Since structs smcd_dmb and ism_dmb are identical, we can simply convert between the two for now. Signed-off-by:
-
Stefan Raspl authored
Signed-off-by:
-
Stefan Raspl authored
Removing an ISM device prior to terminating its associated connections doesn't end well. Signed-off-by:
-
Parav Pandit authored
virtio queue index can be maximum of 65535. 16 bytes are enough to store the vq name with the existing string prefix. With this change, send queue struct saves 24 bytes and receive queue saves whole cache line worth 64 bytes per structure due to saving in alignment bytes. Pahole results before: pahole -s drivers/net/virtio_net.o | \ grep -e "send_queue" -e "receive_queue" send_queue 1112 0 receive_queue 1280 1 Pahole results after: pahole -s drivers/net/virtio_net.o | \ grep -e "send_queue" -e "receive_queue" send_queue 1088 0 receive_queue 1216 1 Signed-off-by:Parav Pandit <parav@nvidia.com> Reviewed-by:
Pavan Chebbi <pavan.chebbi@broadcom.com> Signed-off-by:
-
Jakub Kicinski authored
Build bot detects that err may be returned uninitialized in devlink_fmsg_prepare_skb(). This is not really true because all fmsgs users should create at least one outer nest, and therefore fmsg can't be completely empty. That said the assumption is not trivial to confirm, so let's follow the bots advice, anyway. This code does not seem to have changed since its inception in commit 1db64e87 ("devlink: Add devlink formatted message (fmsg) API") Reviewed-by:
Jiri Pirko <jiri@nvidia.com> Link: https://lore.kernel.org/r/20230124035231.787381-1-kuba@kernel.orgSigned-off-by:
Jakub Kicinski <kuba@kernel.org>
-
Vladimir Oltean authored
We don't read the verify_enabled variable from hardware in the MAC Merge layer state GET operation, instead we always leave it set to "false". The user may think something is wrong if they set verify_enabled to true, then read it back and see it's still false, even though the configuration took place. Fixes: 6505b680 ("net: mscc: ocelot: add MAC Merge layer support for VSC9959") Signed-off-by:
Vladimir Oltean <vladimir.oltean@nxp.com> Link: https://lore.kernel.org/r/20230123184538.3420098-1-vladimir.oltean@nxp.comSigned-off-by:
-
James Hershaw authored
The changes in this patch are as follows: - Alter the logic of get/set_eeprom functions to use the helper function nfp_app_from_netdev() which handles differentiating between an nfp_net and a nfp_repr. This allows us to get an agnostic backpointer to the pdev. - Enable the various eeprom commands by adding the 'get_eeprom_len', 'get_eeprom', 'set_eeprom' callbacks to the nfp_port_ethtool_ops struct. This allows the eeprom commands to work on representor interfaces, similar to a previous patch which added it to the vnics. Currently these are being used to configure persistent MAC addresses for the physical ports on the nfp. Signed-off-by:
James Hershaw <james.hershaw@corigine.com> Reviewed-by:
Louis Peens <louis.peens@corigine.com> Signed-off-by:
Simon Horman <simon.horman@corigine.com> Link: https://lore.kernel.org/r/20230123134135.293278-1-simon.horman@corigine.comSigned-off-by:
-
Guillaume Nault authored
This function is only used in net/ipv6/route.c and has no reason to be visible outside of it. Signed-off-by:
Guillaume Nault <gnault@redhat.com> Reviewed-by:
David Ahern <dsahern@kernel.org> Link: https://lore.kernel.org/r/50706db7f675e40b3594d62011d9363dce32b92e.1674495822.git.gnault@redhat.comSigned-off-by:
-
Jakub Kicinski authored
Commit 2c7bc10d ("netlink: add macro for checking dump ctx size") misspelled the name of the assert as asset, missing an R. Reported-by:
Ido Schimmel <idosch@idosch.org> Reviewed-by:
Ido Schimmel <idosch@nvidia.com> Link: https://lore.kernel.org/r/20230123222224.732338-1-kuba@kernel.orgSigned-off-by:
-
- 24 Jan, 2023 27 commits
-
-
Paolo Abeni authored
Jakub Kicinski says: ==================== Netlink protocol specs I think the Netlink proto specs are far along enough to merge. Filling in all attribute types and quirks will be an ongoing effort but we have enough to cover FOU so it's somewhat complete. I fully intend to continue polishing the code but at the same time I'd like to start helping others base their work on the specs (e.g. DPLL) and need to start working on some new families myself. That's the progress / motivation for merging. The RFC [1] has more of a high level blurb, plus I created a lot of documentation, I'm not going to repeat it here. There was also the talk at LPC [2]. [1] https://lore.kernel.org/all/20220811022304.583300-1-kuba@kernel.org/ [2] https://youtu.be/9QkXIQXkaQk?t=2562 v2: https://lore.kernel.org/all/20220930023418.1346263-1-kuba@kernel.org/ v3: https://lore.kernel.org/all/20230119003613.111778-1-kuba@kernel.org/1 v4: - spec improvements (patch 2) - Python cleanup (patch 3) - rename auto-gen files and use the right comment style ==================== Link: https://lore.kernel.org/r/20230120175041.342573-1-kuba@kernel.orgSigned-off-by:
Paolo Abeni <pabeni@redhat.com>
-
Jakub Kicinski authored
Add a CLI sample which can take in arbitrary request in JSON format, convert it to Netlink and do the inverse for output. It's meant as a development tool primarily and perhaps for selftests which need to tickle netlink in a special way. Acked-by:
Stanislav Fomichev <sdf@google.com> Signed-off-by:
-
Jakub Kicinski authored
Generate and plug in the spec-based tables. A little bit of renaming is needed in the FOU code. Acked-by:
-
Jakub Kicinski authored
We'll need to link two objects together to form the fou module. This means the source can't be called fou, the build system expects fou.o to be the combined object. Acked-by:
-
Jakub Kicinski authored
Regenerate the FOU uAPI header from the YAML spec. The flags now come before attributes which use them, and the comments for type disappear (coders should look at the spec instead). Acked-by:
-
Jakub Kicinski authored
FOU has a reasonably modern Genetlink family. Add a spec. Acked-by:
-
Jakub Kicinski authored
Code generators to turn Netlink specs into C code. I'm definitely not proud of it. The main generator is in Python, there's a bash script to regen all code-gen'ed files in tree after making spec changes. Acked-by:
-
Jakub Kicinski authored
Add schemas for Netlink spec files. As described in the docs we have 4 "protocols" or compatibility levels, and each one comes with its own schema, but the more general / legacy schemas are superset of more modern ones: genetlink is the smallest followed by genetlink-c and genetlink-legacy. There is no schema for raw netlink, yet, I haven't found the time.. I don't know enough jsonschema to do inheritance or something but the repetition is not too bad. I hope. Acked-by:
-
Jakub Kicinski authored
Add documentation about the upcoming Netlink protocol specs. Reviewed-by:
Nicolas Dichtel <nicolas.dichtel@6wind.com> Reviewed-by:
Bagas Sanjaya <bagasdotme@gmail.com> Acked-by:
-
Paolo Abeni authored
Davide Caratti says: ==================== net/sched: use the backlog for nested mirred ingress TC mirred has a protection against excessive stack growth, but that protection doesn't really guarantee the absence of recursion, nor it guards against loops. Patch 1/2 rewords "recursion" to "nesting" to make this more clear. We can leverage on this existing mechanism to prevent TCP / SCTP from doing soft lock-up in some specific scenarios that uses mirred egress->ingress: patch 2 changes mirred so that the networking backlog is used for nested mirred ingress actions. ==================== Link: https://lore.kernel.org/r/cover.1674233458.git.dcaratti@redhat.comSigned-off-by:
-
Davide Caratti authored
William reports kernel soft-lockups on some OVS topologies when TC mirred egress->ingress action is hit by local TCP traffic [1]. The same can also be reproduced with SCTP (thanks Xin for verifying), when client and server reach themselves through mirred egress to ingress, and one of the two peers sends a "heartbeat" packet (from within a timer). Enqueueing to backlog proved to fix this soft lockup; however, as Cong noticed [2], we should preserve - when possible - the current mirred behavior that counts as "overlimits" any eventual packet drop subsequent to the mirred forwarding action [3]. A compromise solution might use the backlog only when tcf_mirred_act() has a nest level greater than one: change tcf_mirred_forward() accordingly. Also, add a kselftest that can reproduce the lockup and verifies TC mirred ability to account for further packet drops after TC mirred egress->ingress (when the nest level is 1). [1] https://lore.kernel.org/netdev/33dc43f587ec1388ba456b4915c75f02a8aae226.1663945716.git.dcaratti@redhat.com/ [2] https://lore.kernel.org/netdev/Y0w%2FWWY60gqrtGLp@pop-os.localdomain/ [3] such behavior is not guaranteed: for example, if RPS or skb RX timestamping is enabled on the mirred target device, the kernel can defer receiving the skb and return NET_RX_SUCCESS inside tcf_mirred_forward(). Reported-by:
William Zhao <wizhao@redhat.com> CC: Xin Long <lucien.xin@gmail.com> Signed-off-by:
Davide Caratti <dcaratti@redhat.com> Reviewed-by:
Marcelo Ricardo Leitner <marcelo.leitner@gmail.com> Acked-by:
Jamal Hadi Salim <jhs@mojatatu.com> Signed-off-by:
-
Davide Caratti authored
with commit e2ca070f ("net: sched: protect against stack overflow in TC act_mirred"), act_mirred protected itself against excessive stack growth using per_cpu counter of nested calls to tcf_mirred_act(), and capping it to MIRRED_RECURSION_LIMIT. However, such protection does not detect recursion/loops in case the packet is enqueued to the backlog (for example, when the mirred target device has RPS or skb timestamping enabled). Change the wording from "recursion" to "nesting" to make it more clear to readers. CC: Jamal Hadi Salim <jhs@mojatatu.com> Signed-off-by:
-
Paolo Abeni authored
Siddharth Vadapalli says: ==================== Fix CPTS release action in am65-cpts driver Delete unreachable code in am65_cpsw_init_cpts() function, which was Reported-by:
Leon Romanovsky <leon@kernel.org> at: https://lore.kernel.org/r/Y8aHwSnVK9+sAb24@unreal Remove the devm action associated with am65_cpts_release() and invoke the function directly on the cleanup and exit paths. v4: https://lore.kernel.org/r/20230120044201.357950-1-s-vadapalli@ti.com/ v3: https://lore.kernel.org/r/20230118095439.114222-1-s-vadapalli@ti.com/ v2: https://lore.kernel.org/r/20230116044517.310461-1-s-vadapalli@ti.com/ v1: https://lore.kernel.org/r/20230113104816.132815-1-s-vadapalli@ti.com/ ==================== Link: https://lore.kernel.org/r/20230120070731.383729-1-s-vadapalli@ti.comSigned-off-by:
-
Siddharth Vadapalli authored
The am65_cpts_release() function is registered as a devm_action in the am65_cpts_create() function in am65-cpts driver. When the am65-cpsw driver invokes am65_cpts_create(), am65_cpts_release() is added in the set of devm actions associated with the am65-cpsw driver's device. In the event of probe failure or probe deferral, the platform_drv_probe() function invokes dev_pm_domain_detach() which powers off the CPSW and the CPSW's CPTS hardware, both of which share the same power domain. Since the am65_cpts_disable() function invoked by the am65_cpts_release() function attempts to reset the CPTS hardware by writing to its registers, the CPTS hardware is assumed to be powered on at this point. However, the hardware is powered off before the devm actions are executed. Fix this by getting rid of the devm action for am65_cpts_release() and invoking it directly on the cleanup and exit paths. Fixes: f6bd5952 ("net: ethernet: ti: introduce am654 common platform time sync driver") Signed-off-by:
Siddharth Vadapalli <s-vadapalli@ti.com> Reviewed-by:
Leon Romanovsky <leonro@nvidia.com> Reviewed-by:
Tony Nguyen <anthony.l.nguyen@intel.com> Reviewed-by:
Roger Quadros <rogerq@kernel.org> Signed-off-by:
-
Siddharth Vadapalli authored
The am65_cpts_create() function returns -EOPNOTSUPP only when the config "CONFIG_TI_K3_AM65_CPTS" is disabled. Also, in the am65_cpsw_init_cpts() function, am65_cpts_create() can only be invoked if the config "CONFIG_TI_K3_AM65_CPTS" is enabled. Thus, the error handling code for the case in which the return value of am65_cpts_create() is -EOPNOTSUPP, is unreachable. Hence delete it. Reported-by:
-
Rakesh Sankaranarayanan authored
PHY initialization is supposed to run on every mode changes. "lan87xx_config_aneg()" verifies every mode change using "phy_modify_changed()" function. Earlier code had phy_modify_changed() followed by genphy_soft_reset. But soft_reset resets all the pre-configured register values to default state, and lost all the initialization done. With this reason gen_phy_reset was removed. But it need to go through init sequence each time the mode changed. Update lan87xx_config_aneg() to invoke phy_init once successful mode update is detected. PHY init sequence added in lan87xx_phy_init() have slave init commands executed every time. Update the init sequence to run slave init only if phydev is in slave mode. Test setup contains LAN9370 EVB connected to SAMA5D3 (Running DSA), and issue can be reproduced by connecting link to any of the available ports after SAMA5D3 boot-up. With this issue, port will fail to update link state. But once the SAMA5D3 is reset with LAN9370 link in connected state itself, on boot-up link state will be reported as UP. But Again after some time, if link is moved to DOWN state, it will not get reported. Signed-off-by:
Rakesh Sankaranarayanan <rakesh.sankaranarayanan@microchip.com> Reviewed-by:
Andrew Lunn <andrew@lunn.ch> Link: https://lore.kernel.org/r/20230120104733.724701-1-rakesh.sankaranarayanan@microchip.comSigned-off-by:
-
Jakub Kicinski authored
Arun Ramadoss says: ==================== net: dsa: microchip: add support for credit based shaper LAN937x switch family, KSZ9477, KSZ9567, KSZ9563 and KSZ8563 supports the credit based shaper. But there were few difference between LAN937x and KSZ switch like - number of queues for LAN937x is 8 and for others it is 4. - size of credit increment register for LAN937x is 24 and for other is 16-bit. This patch series add the credit based shaper with common implementation for LAN937x and KSZ swithes. ==================== Link: https://lore.kernel.org/r/20230120052135.32120-1-arun.ramadoss@microchip.comSigned-off-by:
-
Arun Ramadoss authored
KSZ9477, KSZ9567, KSZ9563, KSZ8563 and LAN937x supports Credit based shaper. To differentiate the chip supporting cbs, tc_cbs_supported flag is introduced in ksz_chip_data. And KSZ series has 16bit Credit increment registers whereas LAN937x has 24bit register. The value to be programmed in the credit increment is determined using the successive multiplication method to convert decimal fraction to hexadecimal fraction. For example: if idleslope is 10000 and sendslope is -90000, then bandwidth is 10000 - (-90000) = 100000. The 10% bandwidth of 100Mbps means 10/100 = 0.1(decimal). This value has to be converted to hexa. 1) 0.1 * 16 = 1.6 --> fraction 0.6 Carry = 1 (MSB) 2) 0.6 * 16 = 9.6 --> fraction 0.6 Carry = 9 3) 0.6 * 16 = 9.6 --> fraction 0.6 Carry = 9 4) 0.6 * 16 = 9.6 --> fraction 0.6 Carry = 9 5) 0.6 * 16 = 9.6 --> fraction 0.6 Carry = 9 6) 0.6 * 16 = 9.6 --> fraction 0.6 Carry = 9 (LSB) Now 0.1(decimal) becomes 0.199999(Hex). If it is LAN937x, 24 bit value will be programmed to Credit Inc register, 0x199999. For others 16 bit value will be prgrammed, 0x1999. Signed-off-by:
Arun Ramadoss <arun.ramadoss@microchip.com> Signed-off-by:
-
Arun Ramadoss authored
LAN937x family of switches has 8 queues per port where the KSZ switches has 4 queues per port. By default, only one queue per port is enabled. The queues are configurable in 2, 4 or 8. This patch add 8 number of queues for LAN937x and 4 for other switches. In the tag_ksz.c file, prioirty of the packet is queried using the skb buffer and the corresponding value is updated in the tag. Signed-off-by:
-
Jesper Dangaard Brouer authored
The spin_lock irqsave/restore API variant in skb_defer_free_flush can be replaced with the faster spin_lock irq variant, which doesn't need to read and restore the CPU flags. Using the unconditional irq "disable/enable" API variant is safe, because the skb_defer_free_flush() function is only called during NAPI-RX processing in net_rx_action(), where it is known the IRQs are enabled. Expected gain is 14 cycles from avoiding reading and restoring CPU flags in a spin_lock_irqsave/restore operation, measured via a microbencmark kernel module[1] on CPU E5-1650 v4 @ 3.60GHz. Microbenchmark overhead of spin_lock+unlock: - spin_lock_unlock_irq cost: 34 cycles(tsc) 9.486 ns - spin_lock_unlock_irqsave cost: 48 cycles(tsc) 13.567 ns We don't expect to see a measurable packet performance gain, as skb_defer_free_flush() is called infrequently once per NIC device NAPI bulk cycle and conditionally only if SKBs have been deferred by other CPUs via skb_attempt_defer_free(). [1] https://github.com/netoptimizer/prototype-kernel/blob/master/kernel/lib/time_bench_sample.cReviewed-by:
Jacob Keller <jacob.e.keller@intel.com> Signed-off-by:
Jesper Dangaard Brouer <brouer@redhat.com> Link: https://lore.kernel.org/r/167421646327.1321776.7390743166998776914.stgit@firesoulSigned-off-by:
-
Jon Maxwell authored
v4: fix deprecated typo. Document that max_size is deprecated due to: commit af6d1034 ("ipv6: remove max_size check inline with ipv4") Signed-off-by:
Jon Maxwell <jmaxwell37@gmail.com> Link: https://lore.kernel.org/r/20230120232331.1273881-1-jmaxwell37@gmail.comSigned-off-by:
-
Jesper Dangaard Brouer authored
A bug was introduced by commit eedade12 ("net: kfree_skb_list use kmem_cache_free_bulk"). It unconditionally unlinked the SKB list via invoking skb_mark_not_on_list(). In this patch we choose to remove the skb_mark_not_on_list() call as it isn't necessary. It would be possible and correct to call skb_mark_not_on_list() only when __kfree_skb_reason() returns true, meaning the SKB is ready to be free'ed, as it calls/check skb_unref(). This fix is needed as kfree_skb_list() is also invoked on skb_shared_info frag_list (skb_drop_fraglist() calling kfree_skb_list()). A frag_list can have SKBs with elevated refcnt due to cloning via skb_clone_fraglist(), which takes a reference on all SKBs in the list. This implies the invariant that all SKBs in the list must have the same refcnt, when using kfree_skb_list(). Reported-by: syzbot+c8a2e66e37eee553c4fd@syzkaller.appspotmail.com Reported-and-tested-by: syzbot+c8a2e66e37eee553c4fd@syzkaller.appspotmail.com Fixes: eedade12 ("net: kfree_skb_list use kmem_cache_free_bulk") Signed-off-by:
Eric Dumazet <edumazet@google.com> Link: https://lore.kernel.org/r/167421088417.1125894.9761158218878962159.stgit@firesoulSigned-off-by:
-
Dan Carpenter authored
The "eport" variable needs to be initialized to NULL for this code to work. Fixes: 814e7693 ("net: microchip: vcap api: Add a storage state to a VCAP rule") Signed-off-by:
Dan Carpenter <error27@gmail.com> Reviewed-by:
Steen Hegelund <Steen.Hegelund@microchip.com> Link: https://lore.kernel.org/r/Y8qbYAb+YSXo1DgR@kiliSigned-off-by:
-
Heiner Kallweit authored
If mdiobus_get_phy() is called with an invalid addr parameter, then the caller has a bug. Print a call trace to help identifying the caller. Suggested-by:
Heiner Kallweit <hkallweit1@gmail.com> Reviewed-by:
-
Jakub Kicinski authored
Merge tag 'wireless-next-2023-01-23' of git://git.kernel.org/pub/scm/linux/kernel/git/wireless/wireless-next Kalle Valo says: ==================== wireless-next patches for v6.3 First set of patches for v6.3. The most important change here is that the old Wireless Extension user space interface is not supported on Wi-Fi 7 devices at all. We also added a warning if anyone with modern drivers (ie. cfg80211 and mac80211 drivers) tries to use Wireless Extensions, everyone should switch to using nl80211 interface instead. Static WEP support is removed, there wasn't any driver using that anyway so there's no user impact. Otherwise it's smaller features and fixes as usual. Note: As mt76 had tricky conflicts due to the fixes in wireless tree, we decided to merge wireless into wireless-next to solve them easily. There should not be any merge problems anymore. Major changes: cfg80211 - remove never used static WEP support - warn if Wireless Extention interface is used with cfg80211/mac80211 drivers - stop supporting Wireless Extensions with Wi-Fi 7 devices - support minimal Wi-Fi 7 Extremely High Throughput (EHT) rate reporting rfkill - add GPIO DT support bitfield - add FIELD_PREP_CONST() mt76 - per-PHY LED support rtw89 - support new Bluetooth co-existance version rtl8xxxu - support RTL8188EU * tag 'wireless-next-2023-01-23' of git://git.kernel.org/pub/scm/linux/kernel/git/wireless/wireless-next: (123 commits) wifi: wireless: deny wireless extensions on MLO-capable devices wifi: wireless: warn on most wireless extension usage wifi: mac80211: drop extra 'e' from ieeee80211... name wifi: cfg80211: Deduplicate certificate loading bitfield: add FIELD_PREP_CONST() wifi: mac80211: add kernel-doc for EHT structure mac80211: support minimal EHT rate reporting on RX wifi: mac80211: Add HE MU-MIMO related flags in ieee80211_bss_conf wifi: mac80211: Add VHT MU-MIMO related flags in ieee80211_bss_conf wifi: cfg80211: Use MLD address to indicate MLD STA disconnection wifi: cfg80211: Support 32 bytes KCK key in GTK rekey offload wifi: cfg80211: Fix extended KCK key length check in nl80211_set_rekey_data() wifi: cfg80211: remove support for static WEP wifi: rtl8xxxu: Dump the efuse only for untested devices wifi: rtl8xxxu: Print the ROM version too wifi: rtw88: Use non-atomic sta iterator in rtw_ra_mask_info_update() wifi: rtw88: Use rtw_iterate_vifs() for rtw_vif_watch_dog_iter() wifi: rtw88: Move register access from rtw_bf_assoc() outside the RCU wifi: rtl8xxxu: Use a longer retry limit of 48 wifi: rtl8xxxu: Report the RSSI to the firmware ... ==================== Link: https://lore.kernel.org/r/20230123103338.330CBC433EF@smtp.kernel.orgSigned-off-by:
-
Krzysztof Kozlowski authored
The AX88796C device node on SPI bus can use SPI peripheral properties in certain configurations: exynos3250-artik5-eval.dtb: ethernet@0: 'controller-data' does not match any of the regexes: 'pinctrl-[0-9]+' Signed-off-by:
Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org> Acked-by:
Rob Herring <robh@kernel.org> Link: https://lore.kernel.org/r/20230120144329.305655-1-krzysztof.kozlowski@linaro.orgSigned-off-by:
-
Eric Dumazet authored
In commit 72653ae5 ("selftests: net: tcp_mmap: Use huge pages in send path") I made a change to use hugepages for the buffer used by the client (tx path) Today, I understood that the cause for poor zerocopy performance was that after a mmap() for a 512KB memory zone, kernel uses a single zeropage, mapped 128 times. This was really the reason for poor tx path performance in zero copy mode, because this zero page refcount is under high pressure, especially when TCP ACK packets are processed on another cpu. We need either to force a COW on all the memory range, or use MAP_POPULATE so that a zero page is not abused. Signed-off-by:
-
- 23 Jan, 2023 1 commit
-
-
Heiner Kallweit authored
Use devm_clk_get_enabled() to simplify the code. Signed-off-by:
Jerome Brunet <jbrunet@baylibre.com> Signed-off-by:
-
