- 14 Dec, 2023 1 commit
-
-
Pedro Tammela authored
Instead of relying only on the idrinfo->lock mutex for bind/alloc logic, rely on a combination of rcu + mutex + atomics to better scale the case where multiple rtnl-less filters are binding to the same action object. Action binding happens when an action index is specified explicitly and an action exists which such index exists. Example: tc actions add action drop index 1 tc filter add ... matchall action drop index 1 tc filter add ... matchall action drop index 1 tc filter add ... matchall action drop index 1 tc filter ls ... filter protocol all pref 49150 matchall chain 0 filter protocol all pref 49150 matchall chain 0 handle 0x1 not_in_hw action order 1: gact action drop random type none pass val 0 index 1 ref 4 bind 3 filter protocol all pref 49151 matchall chain 0 filter protocol all pref 49151 matchall chain 0 handle 0x1 not_in_hw action order 1: gact action drop random type none pass val 0 index 1 ref 4 bind 3 filter protocol all pref 49152 matchall chain 0 filter protocol all pref 49152 matchall chain 0 handle 0x1 not_in_hw action order 1: gact action drop random type none pass val 0 index 1 ref 4 bind 3 When no index is specified, as before, grab the mutex and allocate in the idr the next available id. In this version, as opposed to before, it's simplified to store the -EBUSY pointer instead of the previous alloc + replace combination. When an index is specified, rely on rcu to find if there's an object in such index. If there's none, fallback to the above, serializing on the mutex and reserving the specified id. If there's one, it can be an -EBUSY pointer, in which case we just try again until it's an action, or an action. Given the rcu guarantees, the action found could be dead and therefore we need to bump the refcount if it's not 0, handling the case it's in fact 0. As bind and the action refcount are already atomics, these increments can happen without the mutex protection while many tcf_idr_check_alloc race to bind to the same action instance. In case binding encounters a parallel delete or add, it will return -EAGAIN in order to try again. Both filter and action apis already have the retry machinery in-place. In case it's an unlocked filter it retries under the rtnl lock. Signed-off-by:Pedro Tammela <pctammela@mojatatu.com> Acked-by:
Jamal Hadi Salim <jhs@mojatatu.com> Reviewed-by:
Vlad Buslov <vladbu@nvidia.com> Link: https://lore.kernel.org/r/20231211181807.96028-2-pctammela@mojatatu.comSigned-off-by:
Jakub Kicinski <kuba@kernel.org>
-
- 13 Dec, 2023 39 commits
-
-
David S. Miller authored
Heng Qi says: ==================== virtio-net: support dynamic coalescing moderation Now, virtio-net already supports per-queue moderation parameter setting. Based on this, we use the linux dimlib to support dynamic coalescing moderation for virtio-net. Due to some scheduling issues, we only support and test the rx dim. Some test results: I. Sockperf UDP ================================================= 1. Env rxq_0 with affinity to cpu_0. 2. Cmd client: taskset -c 0 sockperf tp -p 8989 -i $IP -t 10 -m 16B server: taskset -c 0 sockperf sr -p 8989 3. Result dim off: 1143277.00 rxpps, throughput 17.844 MBps, cpu is 100%. dim on: 1124161.00 rxpps, throughput 17.610 MBps, cpu is 83.5%. ================================================= II. Redis ================================================= 1. Env There are 8 rxqs, and rxq_i with affinity to cpu_i. 2. Result When all cpus are 100%, ops/sec of memtier_benchmark client is dim off: 978437.23 dim on: 1143638.28 ================================================= III. Nginx ================================================= 1. Env There are 8 rxqs and rxq_i with affinity to cpu_i. 2. Result When all cpus are 100%, requests/sec of wrk client is dim off: 877931.67 dim on: 1019160.31 ================================================= IV. Latency of sockperf udp ================================================= 1. Rx cmd taskset -c 0 sockperf sr -p 8989 2. Tx cmd taskset -c 0 sockperf pp -i ${ip} -p 8989 -t 10 After running this cmd 5 times and averaging the results, 3. Result dim off: 17.7735 usec dim on: 18.0110 usec ================================================= Changelog: v7->v8: - Add select DIMLIB. v6->v7: - Drop the patch titled "spin lock for ctrl cmd access" - Use rtnl_trylock to avoid the deadlock. v5->v6: - Add patch(4/5): spin lock for ctrl cmd access - Patch(5/5): - Use spin lock and cancel_work_sync to synchronize v4->v5: - Patch(4/4): - Fix possible synchronization issues with cancel_work_sync. - Reduce if/else nesting levels v3->v4: - Patch(5/5): drop. v2->v3: - Patch(4/5): some minor modifications. v1->v2: - Patch(2/5): a minor fix. - Patch(4/5): - improve the judgment of dim switch conditions. - Cancel the work when vq reset. - Patch(5/5): drop the tx dim implementation. ==================== Signed-off-by:David S. Miller <davem@davemloft.net>
-
Heng Qi authored
By comparing the traffic information in the complete napi processes, let the virtio-net driver automatically adjust the coalescing moderation parameters of each receive queue. Signed-off-by:
Heng Qi <hengqi@linux.alibaba.com> Signed-off-by:
-
Heng Qi authored
Extract commands to set virtqueue coalescing parameters for reuse by ethtool -Q, vq resize and netdim. Signed-off-by:
Jason Wang <jasowang@redhat.com> Signed-off-by:
-
Heng Qi authored
This patch separates the rx and tx global coalescing moderation commands to support netdim switches in subsequent patches. Signed-off-by:
-
Heng Qi authored
rx netdim needs to count the traffic during a complete napi process, and start updating and comparing samples to make decisions after the napi ends. Let virtqueue_napi_complete() return true if napi is done, otherwise vice versa. Signed-off-by:
-
David S. Miller authored
Shannon Nelson says: ==================== ionic: updates to PCI error handling These are improvements to our PCI error handling, including FLR and AER events. ==================== Signed-off-by: -
Shannon Nelson authored
Set up the pci_error_handlers error_detected and resume to be useful in handling AER events. If the error detected is pci_channel_io_frozen we set up to do an FLR at the end of the AER handling - this tends to clear things up well enough that traffic can continue. Else, let the AER/PCI machinery do what is needed for the less serious errors seen. Signed-off-by:
Shannon Nelson <shannon.nelson@amd.com> Reviewed-by:
Brett Creeley <brett.creeley@amd.com> Signed-off-by:
-
Shannon Nelson authored
Remove and restore the lif's debugfs pointers on a reset, and make sure to check for the dentry before removing it in case an earlier reset failed to rebuild the lif. Signed-off-by:
-
Shannon Nelson authored
When stopping the watchdog timer at remove time we should be using the new timer_shutdown_sync to assure the timer doesn't ever get rearmed. Signed-off-by:
-
Shannon Nelson authored
If there was a failed attempt to reset the PCI connection, don't later try to read from PCI as the space is unmapped and will cause a paging request crash. When clearing the PCI setup we can clear the dev_info register pointer, and check it before using it in the fw_running test. Signed-off-by:
-
Shannon Nelson authored
If a reset fails, the PCI device is left in a disabled state, so don't try to disable it again on driver remove. This prevents a scary looking WARN trace in the kernel log. ionic 0000:2b:00.0: disabling already-disabled device Signed-off-by: -
Shannon Nelson authored
If the driver or firmware is stuck in reset state, don't bother trying to use adminq commands. This speeds up shutdown and prevents unnecessary timeouts and error messages. This includes a bit of rework on ionic_adminq_post_wait() and ionic_adminq_post_wait_nomsg() to both use __ionic_adminq_post_wait() which can do the checks needed in both cases. Signed-off-by:
-
Shannon Nelson authored
Make sure we keep and replay the filters and RSS config across an FLR by using our FW_RESET flag. This gets checked on the way down and on the way back up to help determine how much LIF state to keep and restore across a reset action. Signed-off-by:
-
Shannon Nelson authored
Don't rely on the PCI memory for the devcmd opcode because we read a 0xff value if the PCI bus is broken, which can cause us to report a bogus dev_cmd opcode later. Signed-off-by:
-
Furong Xu authored
Complete all counters on XGMAC Core. These can be useful for debugging. Signed-off-by:
Furong Xu <0x1207@gmail.com> Signed-off-by:
-
David S. Miller authored
Christian Marangi says: ==================== net: phy: at803x: cleanup The intention of this big series is to try to cleanup the big at803x PHY driver. It currently have 3 different family of PHY in it. at803x, qca83xx and qca808x. The current codebase required lots of cleanup and reworking to make the split possible as currently there is a greater use of adding special function matching the phy_id. This has been reworked to make the function actually generic and make the change only in more specific one. The result is the addition of micro additional function but that is for good as it massively simplify splitting the driver later. Consider that this is all in preparation for the addition of qca807x PHY driver that will also uso some of the functions of at803x. Subsequent series will come with the actual PHY split and other required cleanup. This is only to start the process with minor changes. Changes v4: - Improve at8031_probe function Changes v3: - Add Reviewed-by tag from Andrew - Split patch 10 (at8031 rename) to rename and move Changes v2: - Drop split part due to series too big - Split changes even more - Fix problem pointed out by Russell (flawed reworked function logic) - Add Reviewed-by tag from Andrew - Minor rework to prevent further code duplication for cdt ==================== Signed-off-by: -
Christian Marangi authored
Drop specific PHY ID check for cable test functions for at803x. This is done to make functions more generic. While at it better describe what the functions does by using more symbolic function names. PHYs that requires to set additional reg are moved to specific function calling the more generic one. cdt_start and cdt_wait_for_completion are changed to take an additional arg to pass specific values specific to the PHY. Signed-off-by:
Christian Marangi <ansuelsmth@gmail.com> Signed-off-by:
-
Christian Marangi authored
Move at8035 specific DT parse for clock out frequency to dedicated probe to make at803x probe function more generic. This is to tidy code and no behaviour change are intended. Detection logic is changed, we check if the clk 25m mask is set and if it's not zero, we assume the qca,clk-out-frequency property is set. The property is checked in the generic at803x_parse_dt called by at803x_probe. Signed-off-by:
Andrew Lunn <andrew@lunn.ch> Signed-off-by:
-
Christian Marangi authored
Move at8031 functions in dedicated section with dedicated at8031 parse_dt and probe. Signed-off-by:
-
Christian Marangi authored
Rename at8031 related DT function name to a more specific name referencing they are only related to at8031 and not to the generic at803x PHY family. Signed-off-by:
-
Christian Marangi authored
Move specific at8031 config_intr bits to dedicated function to make at803x_config_initr more generic. This is needed in preparation for PHY driver split as qca8081 share the same function to setup interrupts. Signed-off-by:
-
Christian Marangi authored
Move specific at8031 WOL enable/disable to dedicated function to make at803x_set_wol more generic. This is needed in preparation for PHY driver split as qca8081 share the same function to toggle WOL settings. In this new implementation WOL module in at8031 is enabled after the generic interrupt is setup. This should not cause any problem as the WOL_INT has a separate implementation and only relay on MAC bits. Signed-off-by:
-
Christian Marangi authored
Move specific at8031 config_init to dedicated function to make at803x_config_init more generic and tidy things up. Signed-off-by:
-
Christian Marangi authored
Move specific at8031 probe mode check to dedicated probe to make at803x_probe more generic and keep code tidy. Signed-off-by:
-
Christian Marangi authored
Move specific DT options for at8031 to specific probe to tidy things up and make at803x_parse_dt more generic. Signed-off-by:
-
Christian Marangi authored
Rework qca83xx specific check to dedicated function to tidy things up and drop useless phy_id check. Also drop an useless link_change_notify for QCA8337 as it did nothing an returned early. Signed-off-by:
-
Christian Marangi authored
The function and the struct related to hw_stats were specific to qca83xx PHY but were called following the convention in the driver of calling everything with at803x prefix. To better organize the code, rename these function a more specific name to better describe that they are specific to 83xx PHY family. Signed-off-by:
-
Christian Marangi authored
Move the WOL disable call to specific at8031 probe to make at803x_probe more generic and drop extra check for PHY ID. Keep the same previous behaviour by first calling at803x_probe and then disabling WOL. Signed-off-by:
-
Christian Marangi authored
Fix passing the wrong reference for config_initr on passing the function pointer, drop the wrong & from at803x_config_intr in the PHY struct. Signed-off-by:
-
Jiri Pirko authored
Mode supported is currently reported to the user exactly the same, as the current mode. That's because mode changing is not implemented. Remove the leftover mode_supported() op and use mode_get() to fill up the supported mode exposed to user. One, if even, mode changing is going to be introduced, this could be very easily taken back. In the meantime, prevent drivers form implementing this in wrong way (as for example recent netdevsim implementation attempt intended to do). Signed-off-by:
Jiri Pirko <jiri@nvidia.com> Reviewed-by:
Simon Horman <horms@kernel.org> Signed-off-by:
-
Jakub Kicinski authored
Michael Chan says: ==================== bnxt_en: Update for net-next The first 4 patches in the series fix issues in the net-next tree introduced in the last 4 weeks. The first 3 patches fix ring accounting and indexing logic. The 4th patch fix TX timeout when the TX ring is very small. The next 7 patches add new features on the P7 chips, including TX coalesced completions, VXLAN GPE and UDP GSO stateless offload, a new rx_filter_miss counters, and more QP backing store memory for RoCE. The last 2 patches are PTP improvements. ==================== Link: https://lore.kernel.org/r/20231212005122.2401-1-michael.chan@broadcom.comSigned-off-by:
-
Pavan Chebbi authored
In a busy network, especially with flow control enabled, we may experience timestamp query failures fairly regularly. After a while, dmesg may be flooded with timestamp query failure error messages. Silence the error message from the low level hwrm function that sends the firmware message. Change netdev_err() to netdev_WARN_ONCE() if this FW call ever fails. Signed-off-by:
Pavan Chebbi <pavan.chebbi@broadcom.com> Signed-off-by:
Michael Chan <michael.chan@broadcom.com> Link: https://lore.kernel.org/r/20231212005122.2401-14-michael.chan@broadcom.comSigned-off-by:
-
Pavan Chebbi authored
We don't have to close and open the nic to make sure we have valid rx timestamps. Once we have the timestamp filter applied to the HW and the timestamp_fld_format bit is cleared in the rx completion and the timestamp is non-zero, we can be sure that rx timestamp is valid data. Skip close/open when we set any timestamp filter. Reviewed-by:
Andy Gospodarek <andrew.gospodarek@broadcom.com> Signed-off-by:
-
Michael Chan authored
The new 5760X chips supports UDP GSO. Tested using udpgso_bench_tx. Reviewed-by:
Somnath Kotur <somnath.kotur@broadcom.com> Signed-off-by:
-
Damodharam Ammepalli authored
rx_filter_miss counter is newly added to the rx_port_stats_ext stats structure for newer chips. Newer firmware will return the structure size that includes this counter. Add this entry to the bnxt_port_stats_ext_arr array and the ethtool -S code will pick up this counter if it is supported. Signed-off-by:
Damodharam Ammepalli <damodharam.ammepalli@broadcom.com> Signed-off-by:
-
Michael Chan authored
On the new P7 chips, TPA for tunnel packets can be independently enabled for each VNIC. The default TPA configuration should not include UDP tunnels because the UDP ports for these tunnels are not known yet. The chip should not aggregate these UDP tunneled packets using default UDP ports until the ports are known. Add a new function bnxt_hwrm_vnic_update_tunl_tpa() to enable VXLAN and Geneve TPA if the corresponding UDP ports are known. Reviewed-by:
Ajit Khaparde <ajit.khaparde@broadcom.com> Signed-off-by:
-
Michael Chan authored
Add a new bnxt_udp_tunnels_p7 struct to support the new P7 chips that can parse VXLAN GPE packets. Add VXLAN GPE tunnel type handling to the .set_port() and .unset_port() functions. .ndo_features_check() is also enhanced to support VXLAN GPE which may encapsulate inner IP packets instead of ethernet packets. Reviewed-by:
-
Michael Chan authored
In bnxt_udp_tunnel_set_port(), use the proper ALLOC commands instead of the FREE commands for correctness. The ALLOC and FREE commands happen to be identical so this is just a cosmetic fix for correctness. Reviewed-by:
-
Selvin Xavier authored
The Fast QP modify destroy RoCE feature requires additional QP entries in QP context backing store. FW reports the extra count to be allocated during backing store query. Use this value and allocate extra memory. Note that this works for both the V1 and V1 backing store FW APIs. Signed-off-by:
Selvin Xavier <selvin.xavier@broadcom.com> Signed-off-by:
-
