kernel.git/drivers/net/ethernet/intel, branch linux-5.19.y

iavf: Fix race between iavf_close and iavf_reset_task

2022-10-24T07:58:13Z

[ Upstream commit 11c12adcbc1598d91e73ab6ddfa41d25a01478ed ] During stress tests with adding VF to namespace and changing vf's trust there was a race between iavf_reset_task and iavf_close. Sometimes when IAVF_FLAG_AQ_DISABLE_QUEUES from iavf_close was sent to PF after reset and before IAVF_AQ_GET_CONFIG was sent then PF returns error IAVF_NOT_SUPPORTED to disable queues request and following requests. There is need to get_config before other aq_required will be send but iavf_close clears all flags, if get_config was not sent before iavf_close, then it will not be send at all. In case when IAVF_FLAG_AQ_GET_OFFLOAD_VLAN_V2_CAPS was sent before IAVF_FLAG_AQ_DISABLE_QUEUES then there was rtnl_lock deadlock between iavf_close and iavf_adminq_task until iavf_close timeouts and disable queues was sent after iavf_close ends. There was also a problem with sending delete/add filters. Sometimes when filters was not yet added to PF and in iavf_close all filters was set to remove there might be a try to remove nonexistent filters on PF. Add aq_required_tmp to save aq_required flags and send them after disable_queues will be handled. Clear flags given to iavf_down different than IAVF_FLAG_AQ_GET_CONFIG as this flag is necessary to sent other aq_required. Remove some flags that we don't want to send as we are in iavf_close and we want to disable interface. Remove filters which was not yet sent and send del filters flags only when there are filters to remove. Signed-off-by: Michal Jaron Signed-off-by: Mateusz Palczewski Tested-by: Konrad Jankowski Signed-off-by: Tony Nguyen Signed-off-by: Sasha Levin

ice: set tx_tstamps when creating new Tx rings via ethtool

2022-10-24T07:58:11Z

[ Upstream commit b3b173745c8cab1e24d6821488b60abed3acb24d ] When the user changes the number of queues via ethtool, the driver allocates new rings. This allocation did not initialize tx_tstamps. This results in the tx_tstamps field being zero (due to kcalloc allocation), and would result in a NULL pointer dereference when attempting a transmit timestamp on the new ring. Signed-off-by: Jacob Keller Tested-by: Gurucharan (A Contingent worker at Intel) Signed-off-by: Tony Nguyen Signed-off-by: Sasha Levin

ice: xsk: drop power of 2 ring size restriction for AF_XDP

2022-10-05T08:40:53Z

[ Upstream commit b3056ae2b57858b02b376b3fed6077040caf14b4 ] We had multiple customers in the past months that reported commit 296f13ff3854 ("ice: xsk: Force rings to be sized to power of 2") makes them unable to use ring size of 8160 in conjunction with AF_XDP. Remove this restriction. Fixes: 296f13ff3854 ("ice: xsk: Force rings to be sized to power of 2") CC: Alasdair McWilliam Signed-off-by: Maciej Fijalkowski Tested-by: George Kuruvinakunnel Signed-off-by: Tony Nguyen Signed-off-by: Sasha Levin

ice: xsk: change batched Tx descriptor cleaning

2022-10-05T08:40:52Z

[ Upstream commit 29322791bc8b4f42fc65734840826e3ddc30921e ] AF_XDP Tx descriptor cleaning in ice driver currently works in a "lazy" way - descriptors are not cleaned immediately after send. We rather hold on with cleaning until we see that free space in ring drops below particular threshold. This was supposed to reduce the amount of unnecessary work related to cleaning and instead of keeping the ring empty, ring was rather saturated. In AF_XDP realm cleaning Tx descriptors implies producing them to CQ. This is a way of letting know user space that particular descriptor has been sent, as John points out in [0]. We tried to implement serial descriptor cleaning which would be used in conjunction with batched cleaning but it made code base more convoluted and probably harder to maintain in future. Therefore we step away from batched cleaning in a current form in favor of an approach where we set RS bit on every last descriptor from a batch and clean always at the beginning of ice_xmit_zc(). This means that we give up a bit of Tx performance, but this doesn't hurt l2fwd scenario which is way more meaningful than txonly as this can be treaten as AF_XDP based packet generator. l2fwd is not hurt due to the fact that Tx side is much faster than Rx and Rx is the one that has to catch Tx up. FWIW Tx descriptors are still produced in a batched way. [0]: https://lore.kernel.org/bpf/62b0a20232920_3573208ab@john.notmuch/ Fixes: 126cdfe1007a ("ice: xsk: Improve AF_XDP ZC Tx and use batching API") Signed-off-by: Maciej Fijalkowski Tested-by: George Kuruvinakunnel Signed-off-by: Tony Nguyen Signed-off-by: Sasha Levin

ice: Fix ice_xdp_xmit() when XDP TX queue number is not sufficient

2022-09-28T09:32:20Z

[ Upstream commit 114f398d48c571bb628187a7b2dd42695156781f ] The original patch added the static branch to handle the situation, when assigning an XDP TX queue to every CPU is not possible, so they have to be shared. However, in the XDP transmit handler ice_xdp_xmit(), an error was returned in such cases even before static condition was checked, thus making queue sharing still impossible. Fixes: 22bf877e528f ("ice: introduce XDP_TX fallback path") Signed-off-by: Larysa Zaremba Reviewed-by: Alexander Lobakin Link: https://lore.kernel.org/r/20220919134346.25030-1-larysa.zaremba@intel.com Signed-off-by: Jakub Kicinski Signed-off-by: Sasha Levin

ice: Fix interface being down after reset with link-down-on-close flag on

2022-09-28T09:32:18Z

[ Upstream commit 8ac7132704f3fbd2095abb9459e5303ce8c9e559 ] When performing a reset on ice driver with link-down-on-close flag on interface would always stay down. Fix this by moving a check of this flag to ice_stop() that is called only when user wants to bring interface down. Fixes: ab4ab73fc1ec ("ice: Add ethtool private flag to make forcing link down optional") Signed-off-by: Mateusz Palczewski Tested-by: Petr Oros Tested-by: Gurucharan (A Contingent worker at Intel) Signed-off-by: Tony Nguyen Signed-off-by: Sasha Levin

ice: config netdev tc before setting queues number

2022-09-28T09:32:18Z

[ Upstream commit 122045ca770492660005c22379995506f13efea8 ] After lowering number of tx queues the warning appears: "Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!" Example command to reproduce: ethtool -L enp24s0f0 tx 36 rx 36 Fix this by setting correct tc mapping before setting real number of queues on netdev. Fixes: 0754d65bd4be5 ("ice: Add infrastructure for mqprio support via ndo_setup_tc") Signed-off-by: Michal Swiatkowski Tested-by: Gurucharan (A Contingent worker at Intel) Signed-off-by: Tony Nguyen Signed-off-by: Sasha Levin

i40e: Fix set max_tx_rate when it is lower than 1 Mbps

2022-09-28T09:32:15Z

[ Upstream commit 198eb7e1b81d8ba676d0f4f120c092032ae69a8e ] While converting max_tx_rate from bytes to Mbps, this value was set to 0, if the original value was lower than 125000 bytes (1 Mbps). This would cause no transmission rate limiting to occur. This happened due to lack of check of max_tx_rate against the 1 Mbps value for max_tx_rate and the following division by 125000. Fix this issue by adding a helper i40e_bw_bytes_to_mbits() which sets max_tx_rate to minimum usable value of 50 Mbps, if its value is less than 1 Mbps, otherwise do the required conversion by dividing by 125000. Fixes: 5ecae4120a6b ("i40e: Refactor VF BW rate limiting") Signed-off-by: Michal Jaron Signed-off-by: Andrii Staikov Tested-by: Bharathi Sreenivas Signed-off-by: Tony Nguyen Signed-off-by: Sasha Levin

i40e: Fix VF set max MTU size

2022-09-28T09:32:15Z

[ Upstream commit 372539def2824c43b6afe2403045b140f65c5acc ] Max MTU sent to VF is set to 0 during memory allocation. It cause that max MTU on VF is changed to IAVF_MAX_RXBUFFER and does not depend on data from HW. Set max_mtu field in virtchnl_vf_resource struct to inform VF in GET_VF_RESOURCES msg what size should be max frame. Fixes: dab86afdbbd1 ("i40e/i40evf: Change the way we limit the maximum frame size for Rx") Signed-off-by: Michal Jaron Signed-off-by: Mateusz Palczewski Tested-by: Konrad Jankowski Signed-off-by: Tony Nguyen Signed-off-by: Sasha Levin

iavf: Fix set max MTU size with port VLAN and jumbo frames

2022-09-28T09:32:15Z

[ Upstream commit 399c98c4dc50b7eb7e9f24da7ffdda6f025676ef ] After setting port VLAN and MTU to 9000 on VF with ice driver there was an iavf error "PF returned error -5 (IAVF_ERR_PARAM) to our request 6". During queue configuration, VF's max packet size was set to IAVF_MAX_RXBUFFER but on ice max frame size was smaller by VLAN_HLEN due to making some space for port VLAN as VF is not aware whether it's in a port VLAN. This mismatch in sizes caused ice to reject queue configuration with ERR_PARAM error. Proper max_mtu is sent from ice PF to VF with GET_VF_RESOURCES msg but VF does not look at this. In iavf change max_frame from IAVF_MAX_RXBUFFER to max_mtu received from pf with GET_VF_RESOURCES msg to make vf's max_frame_size dependent from pf. Add check if received max_mtu is not in eligible range then set it to IAVF_MAX_RXBUFFER. Fixes: dab86afdbbd1 ("i40e/i40evf: Change the way we limit the maximum frame size for Rx") Signed-off-by: Michal Jaron Signed-off-by: Mateusz Palczewski Tested-by: Konrad Jankowski Signed-off-by: Tony Nguyen Signed-off-by: Sasha Levin