From 9b13bd53134c9ddd544a790125199fdbdb505e67 Mon Sep 17 00:00:00 2001 From: Jedrzej Jagielski Date: Thu, 28 Oct 2021 13:51:14 +0000 Subject: [PATCH 1/5] i40e: Increase delay to 1 s after global EMP reset Recently simplified i40e_rebuild causes that FW sometimes is not ready after NVM update, the ping does not return. Increase the delay in case of EMP reset. Old delay of 300 ms was introduced for specific cards for 710 series. Now it works for all the cards and delay was increased. Fixes: 1fa51a650e1d ("i40e: Add delay after EMP reset for firmware to recover") Signed-off-by: Arkadiusz Kubalewski Signed-off-by: Jedrzej Jagielski Tested-by: Gurucharan G Signed-off-by: Tony Nguyen --- drivers/net/ethernet/intel/i40e/i40e_main.c | 12 +++--------- 1 file changed, 3 insertions(+), 9 deletions(-) diff --git a/drivers/net/ethernet/intel/i40e/i40e_main.c b/drivers/net/ethernet/intel/i40e/i40e_main.c index 2a3d8aef7f4e..1b0fd3eacd64 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_main.c +++ b/drivers/net/ethernet/intel/i40e/i40e_main.c @@ -10574,15 +10574,9 @@ static void i40e_rebuild(struct i40e_pf *pf, bool reinit, bool lock_acquired) } i40e_get_oem_version(&pf->hw); - if (test_bit(__I40E_EMP_RESET_INTR_RECEIVED, pf->state) && - ((hw->aq.fw_maj_ver == 4 && hw->aq.fw_min_ver <= 33) || - hw->aq.fw_maj_ver < 4) && hw->mac.type == I40E_MAC_XL710) { - /* The following delay is necessary for 4.33 firmware and older - * to recover after EMP reset. 200 ms should suffice but we - * put here 300 ms to be sure that FW is ready to operate - * after reset. - */ - mdelay(300); + if (test_and_clear_bit(__I40E_EMP_RESET_INTR_RECEIVED, pf->state)) { + /* The following delay is necessary for firmware update. */ + mdelay(1000); } /* re-verify the eeprom if we just had an EMP reset */ From d701658a50a471591094b3eb3961b4926cc8f104 Mon Sep 17 00:00:00 2001 From: Jedrzej Jagielski Date: Fri, 5 Nov 2021 11:17:00 +0000 Subject: [PATCH 2/5] i40e: Fix issue when maximum queues is exceeded Before this patch VF interface vanished when maximum queue number was exceeded. Driver tried to add next queues even if there was not enough space. PF sent incorrect number of queues to the VF when there were not enough of them. Add an additional condition introduced to check available space in 'qp_pile' before proceeding. This condition makes it impossible to add queues if they number is greater than the number resulting from available space. Also add the search for free space in PF queue pair piles. Without this patch VF interfaces are not seen when available space for queues has been exceeded and following logs appears permanently in dmesg: "Unable to get VF config (-32)". "VF 62 failed opcode 3, retval: -5" "Unable to get VF config due to PF error condition, not retrying" Fixes: 7daa6bf3294e ("i40e: driver core headers") Fixes: 41c445ff0f48 ("i40e: main driver core") Signed-off-by: Jaroslaw Gawin Signed-off-by: Slawomir Laba Signed-off-by: Jedrzej Jagielski Tested-by: Konrad Jankowski Signed-off-by: Tony Nguyen --- drivers/net/ethernet/intel/i40e/i40e.h | 1 - drivers/net/ethernet/intel/i40e/i40e_main.c | 14 +---- .../ethernet/intel/i40e/i40e_virtchnl_pf.c | 59 +++++++++++++++++++ 3 files changed, 61 insertions(+), 13 deletions(-) diff --git a/drivers/net/ethernet/intel/i40e/i40e.h b/drivers/net/ethernet/intel/i40e/i40e.h index 4d939af0a626..c8cfe62d5e05 100644 --- a/drivers/net/ethernet/intel/i40e/i40e.h +++ b/drivers/net/ethernet/intel/i40e/i40e.h @@ -174,7 +174,6 @@ enum i40e_interrupt_policy { struct i40e_lump_tracking { u16 num_entries; - u16 search_hint; u16 list[0]; #define I40E_PILE_VALID_BIT 0x8000 #define I40E_IWARP_IRQ_PILE_ID (I40E_PILE_VALID_BIT - 2) diff --git a/drivers/net/ethernet/intel/i40e/i40e_main.c b/drivers/net/ethernet/intel/i40e/i40e_main.c index 1b0fd3eacd64..9456641cd24a 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_main.c +++ b/drivers/net/ethernet/intel/i40e/i40e_main.c @@ -196,10 +196,6 @@ int i40e_free_virt_mem_d(struct i40e_hw *hw, struct i40e_virt_mem *mem) * @id: an owner id to stick on the items assigned * * Returns the base item index of the lump, or negative for error - * - * The search_hint trick and lack of advanced fit-finding only work - * because we're highly likely to have all the same size lump requests. - * Linear search time and any fragmentation should be minimal. **/ static int i40e_get_lump(struct i40e_pf *pf, struct i40e_lump_tracking *pile, u16 needed, u16 id) @@ -214,8 +210,7 @@ static int i40e_get_lump(struct i40e_pf *pf, struct i40e_lump_tracking *pile, return -EINVAL; } - /* start the linear search with an imperfect hint */ - i = pile->search_hint; + i = 0; while (i < pile->num_entries) { /* skip already allocated entries */ if (pile->list[i] & I40E_PILE_VALID_BIT) { @@ -234,7 +229,6 @@ static int i40e_get_lump(struct i40e_pf *pf, struct i40e_lump_tracking *pile, for (j = 0; j < needed; j++) pile->list[i+j] = id | I40E_PILE_VALID_BIT; ret = i; - pile->search_hint = i + j; break; } @@ -257,7 +251,7 @@ static int i40e_put_lump(struct i40e_lump_tracking *pile, u16 index, u16 id) { int valid_id = (id | I40E_PILE_VALID_BIT); int count = 0; - int i; + u16 i; if (!pile || index >= pile->num_entries) return -EINVAL; @@ -269,8 +263,6 @@ static int i40e_put_lump(struct i40e_lump_tracking *pile, u16 index, u16 id) count++; } - if (count && index < pile->search_hint) - pile->search_hint = index; return count; } @@ -11786,7 +11778,6 @@ static int i40e_init_interrupt_scheme(struct i40e_pf *pf) return -ENOMEM; pf->irq_pile->num_entries = vectors; - pf->irq_pile->search_hint = 0; /* track first vector for misc interrupts, ignore return */ (void)i40e_get_lump(pf, pf->irq_pile, 1, I40E_PILE_VALID_BIT - 1); @@ -12589,7 +12580,6 @@ static int i40e_sw_init(struct i40e_pf *pf) goto sw_init_done; } pf->qp_pile->num_entries = pf->hw.func_caps.num_tx_qp; - pf->qp_pile->search_hint = 0; pf->tx_timeout_recovery_level = 1; diff --git a/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c b/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c index b785d09c19f8..7dd2a2bab339 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c +++ b/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c @@ -2617,6 +2617,59 @@ static int i40e_vc_disable_queues_msg(struct i40e_vf *vf, u8 *msg) aq_ret); } +/** + * i40e_check_enough_queue - find big enough queue number + * @vf: pointer to the VF info + * @needed: the number of items needed + * + * Returns the base item index of the queue, or negative for error + **/ +static int i40e_check_enough_queue(struct i40e_vf *vf, u16 needed) +{ + unsigned int i, cur_queues, more, pool_size; + struct i40e_lump_tracking *pile; + struct i40e_pf *pf = vf->pf; + struct i40e_vsi *vsi; + + vsi = pf->vsi[vf->lan_vsi_idx]; + cur_queues = vsi->alloc_queue_pairs; + + /* if current allocated queues are enough for need */ + if (cur_queues >= needed) + return vsi->base_queue; + + pile = pf->qp_pile; + if (cur_queues > 0) { + /* if the allocated queues are not zero + * just check if there are enough queues for more + * behind the allocated queues. + */ + more = needed - cur_queues; + for (i = vsi->base_queue + cur_queues; + i < pile->num_entries; i++) { + if (pile->list[i] & I40E_PILE_VALID_BIT) + break; + + if (more-- == 1) + /* there is enough */ + return vsi->base_queue; + } + } + + pool_size = 0; + for (i = 0; i < pile->num_entries; i++) { + if (pile->list[i] & I40E_PILE_VALID_BIT) { + pool_size = 0; + continue; + } + if (needed <= ++pool_size) + /* there is enough */ + return i; + } + + return -ENOMEM; +} + /** * i40e_vc_request_queues_msg * @vf: pointer to the VF info @@ -2651,6 +2704,12 @@ static int i40e_vc_request_queues_msg(struct i40e_vf *vf, u8 *msg) req_pairs - cur_pairs, pf->queues_left); vfres->num_queue_pairs = pf->queues_left + cur_pairs; + } else if (i40e_check_enough_queue(vf, req_pairs) < 0) { + dev_warn(&pf->pdev->dev, + "VF %d requested %d more queues, but there is not enough for it.\n", + vf->vf_id, + req_pairs - cur_pairs); + vfres->num_queue_pairs = cur_pairs; } else { /* successful request */ vf->num_req_queues = req_pairs; From 92947844b8beee988c0ce17082b705c2f75f0742 Mon Sep 17 00:00:00 2001 From: Sylwester Dziedziuch Date: Fri, 26 Nov 2021 11:11:22 +0100 Subject: [PATCH 3/5] i40e: Fix queues reservation for XDP When XDP was configured on a system with large number of CPUs and X722 NIC there was a call trace with NULL pointer dereference. i40e 0000:87:00.0: failed to get tracking for 256 queues for VSI 0 err -12 i40e 0000:87:00.0: setup of MAIN VSI failed BUG: kernel NULL pointer dereference, address: 0000000000000000 RIP: 0010:i40e_xdp+0xea/0x1b0 [i40e] Call Trace: ? i40e_reconfig_rss_queues+0x130/0x130 [i40e] dev_xdp_install+0x61/0xe0 dev_xdp_attach+0x18a/0x4c0 dev_change_xdp_fd+0x1e6/0x220 do_setlink+0x616/0x1030 ? ahci_port_stop+0x80/0x80 ? ata_qc_issue+0x107/0x1e0 ? lock_timer_base+0x61/0x80 ? __mod_timer+0x202/0x380 rtnl_setlink+0xe5/0x170 ? bpf_lsm_binder_transaction+0x10/0x10 ? security_capable+0x36/0x50 rtnetlink_rcv_msg+0x121/0x350 ? rtnl_calcit.isra.0+0x100/0x100 netlink_rcv_skb+0x50/0xf0 netlink_unicast+0x1d3/0x2a0 netlink_sendmsg+0x22a/0x440 sock_sendmsg+0x5e/0x60 __sys_sendto+0xf0/0x160 ? __sys_getsockname+0x7e/0xc0 ? _copy_from_user+0x3c/0x80 ? __sys_setsockopt+0xc8/0x1a0 __x64_sys_sendto+0x20/0x30 do_syscall_64+0x33/0x40 entry_SYSCALL_64_after_hwframe+0x44/0xae RIP: 0033:0x7f83fa7a39e0 This was caused by PF queue pile fragmentation due to flow director VSI queue being placed right after main VSI. Because of this main VSI was not able to resize its queue allocation for XDP resulting in no queues allocated for main VSI when XDP was turned on. Fix this by always allocating last queue in PF queue pile for a flow director VSI. Fixes: 41c445ff0f48 ("i40e: main driver core") Fixes: 74608d17fe29 ("i40e: add support for XDP_TX action") Signed-off-by: Sylwester Dziedziuch Signed-off-by: Mateusz Palczewski Reviewed-by: Maciej Fijalkowski Tested-by: Kiran Bhandare Signed-off-by: Tony Nguyen --- drivers/net/ethernet/intel/i40e/i40e_main.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+) diff --git a/drivers/net/ethernet/intel/i40e/i40e_main.c b/drivers/net/ethernet/intel/i40e/i40e_main.c index 9456641cd24a..9b65cb50282c 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_main.c +++ b/drivers/net/ethernet/intel/i40e/i40e_main.c @@ -210,6 +210,20 @@ static int i40e_get_lump(struct i40e_pf *pf, struct i40e_lump_tracking *pile, return -EINVAL; } + /* Allocate last queue in the pile for FDIR VSI queue + * so it doesn't fragment the qp_pile + */ + if (pile == pf->qp_pile && pf->vsi[id]->type == I40E_VSI_FDIR) { + if (pile->list[pile->num_entries - 1] & I40E_PILE_VALID_BIT) { + dev_err(&pf->pdev->dev, + "Cannot allocate queue %d for I40E_VSI_FDIR\n", + pile->num_entries - 1); + return -ENOMEM; + } + pile->list[pile->num_entries - 1] = id | I40E_PILE_VALID_BIT; + return pile->num_entries - 1; + } + i = 0; while (i < pile->num_entries) { /* skip already allocated entries */ From 0f344c8129a5337dae50e31b817dd50a60ff238c Mon Sep 17 00:00:00 2001 From: Karen Sornek Date: Thu, 2 Dec 2021 12:52:01 +0100 Subject: [PATCH 4/5] i40e: Fix for failed to init adminq while VF reset Fix for failed to init adminq: -53 while VF is resetting via MAC address changing procedure. Added sync module to avoid reading deadbeef value in reinit adminq during software reset. Without this patch it is possible to trigger VF reset procedure during reinit adminq. This resulted in an incorrect reading of value from the AQP registers and generated the -53 error. Fixes: 5c3c48ac6bf5 ("i40e: implement virtual device interface") Signed-off-by: Grzegorz Szczurek Signed-off-by: Karen Sornek Tested-by: Konrad Jankowski Signed-off-by: Tony Nguyen --- .../net/ethernet/intel/i40e/i40e_register.h | 3 ++ .../ethernet/intel/i40e/i40e_virtchnl_pf.c | 44 ++++++++++++++++++- .../ethernet/intel/i40e/i40e_virtchnl_pf.h | 1 + 3 files changed, 46 insertions(+), 2 deletions(-) diff --git a/drivers/net/ethernet/intel/i40e/i40e_register.h b/drivers/net/ethernet/intel/i40e/i40e_register.h index 8d0588a27a05..1908eed4fa5e 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_register.h +++ b/drivers/net/ethernet/intel/i40e/i40e_register.h @@ -413,6 +413,9 @@ #define I40E_VFINT_DYN_CTLN(_INTVF) (0x00024800 + ((_INTVF) * 4)) /* _i=0...511 */ /* Reset: VFR */ #define I40E_VFINT_DYN_CTLN_CLEARPBA_SHIFT 1 #define I40E_VFINT_DYN_CTLN_CLEARPBA_MASK I40E_MASK(0x1, I40E_VFINT_DYN_CTLN_CLEARPBA_SHIFT) +#define I40E_VFINT_ICR0_ADMINQ_SHIFT 30 +#define I40E_VFINT_ICR0_ADMINQ_MASK I40E_MASK(0x1, I40E_VFINT_ICR0_ADMINQ_SHIFT) +#define I40E_VFINT_ICR0_ENA(_VF) (0x0002C000 + ((_VF) * 4)) /* _i=0...127 */ /* Reset: CORER */ #define I40E_VPINT_AEQCTL(_VF) (0x0002B800 + ((_VF) * 4)) /* _i=0...127 */ /* Reset: CORER */ #define I40E_VPINT_AEQCTL_MSIX_INDX_SHIFT 0 #define I40E_VPINT_AEQCTL_ITR_INDX_SHIFT 11 diff --git a/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c b/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c index 7dd2a2bab339..dfdb6e786461 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c +++ b/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.c @@ -1376,6 +1376,32 @@ static i40e_status i40e_config_vf_promiscuous_mode(struct i40e_vf *vf, return aq_ret; } +/** + * i40e_sync_vfr_reset + * @hw: pointer to hw struct + * @vf_id: VF identifier + * + * Before trigger hardware reset, we need to know if no other process has + * reserved the hardware for any reset operations. This check is done by + * examining the status of the RSTAT1 register used to signal the reset. + **/ +static int i40e_sync_vfr_reset(struct i40e_hw *hw, int vf_id) +{ + u32 reg; + int i; + + for (i = 0; i < I40E_VFR_WAIT_COUNT; i++) { + reg = rd32(hw, I40E_VFINT_ICR0_ENA(vf_id)) & + I40E_VFINT_ICR0_ADMINQ_MASK; + if (reg) + return 0; + + usleep_range(100, 200); + } + + return -EAGAIN; +} + /** * i40e_trigger_vf_reset * @vf: pointer to the VF structure @@ -1390,9 +1416,11 @@ static void i40e_trigger_vf_reset(struct i40e_vf *vf, bool flr) struct i40e_pf *pf = vf->pf; struct i40e_hw *hw = &pf->hw; u32 reg, reg_idx, bit_idx; + bool vf_active; + u32 radq; /* warn the VF */ - clear_bit(I40E_VF_STATE_ACTIVE, &vf->vf_states); + vf_active = test_and_clear_bit(I40E_VF_STATE_ACTIVE, &vf->vf_states); /* Disable VF's configuration API during reset. The flag is re-enabled * in i40e_alloc_vf_res(), when it's safe again to access VF's VSI. @@ -1406,7 +1434,19 @@ static void i40e_trigger_vf_reset(struct i40e_vf *vf, bool flr) * just need to clean up, so don't hit the VFRTRIG register. */ if (!flr) { - /* reset VF using VPGEN_VFRTRIG reg */ + /* Sync VFR reset before trigger next one */ + radq = rd32(hw, I40E_VFINT_ICR0_ENA(vf->vf_id)) & + I40E_VFINT_ICR0_ADMINQ_MASK; + if (vf_active && !radq) + /* waiting for finish reset by virtual driver */ + if (i40e_sync_vfr_reset(hw, vf->vf_id)) + dev_info(&pf->pdev->dev, + "Reset VF %d never finished\n", + vf->vf_id); + + /* Reset VF using VPGEN_VFRTRIG reg. It is also setting + * in progress state in rstat1 register. + */ reg = rd32(hw, I40E_VPGEN_VFRTRIG(vf->vf_id)); reg |= I40E_VPGEN_VFRTRIG_VFSWR_MASK; wr32(hw, I40E_VPGEN_VFRTRIG(vf->vf_id), reg); diff --git a/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.h b/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.h index 49575a640a84..03c42fd0fea1 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.h +++ b/drivers/net/ethernet/intel/i40e/i40e_virtchnl_pf.h @@ -19,6 +19,7 @@ #define I40E_MAX_VF_PROMISC_FLAGS 3 #define I40E_VF_STATE_WAIT_COUNT 20 +#define I40E_VFR_WAIT_COUNT 100 /* Various queue ctrls */ enum i40e_queue_ctrl { From 3b8428b84539c78fdc8006c17ebd25afd4722d51 Mon Sep 17 00:00:00 2001 From: Joe Damato Date: Wed, 8 Dec 2021 17:56:33 -0800 Subject: [PATCH 5/5] i40e: fix unsigned stat widths Change i40e_update_vsi_stats and struct i40e_vsi to use u64 fields to match the width of the stats counters in struct i40e_rx_queue_stats. Update debugfs code to use the correct format specifier for u64. Fixes: 41c445ff0f48 ("i40e: main driver core") Signed-off-by: Joe Damato Reported-by: kernel test robot Tested-by: Gurucharan G Signed-off-by: Tony Nguyen --- drivers/net/ethernet/intel/i40e/i40e.h | 8 ++++---- drivers/net/ethernet/intel/i40e/i40e_debugfs.c | 2 +- drivers/net/ethernet/intel/i40e/i40e_main.c | 4 ++-- 3 files changed, 7 insertions(+), 7 deletions(-) diff --git a/drivers/net/ethernet/intel/i40e/i40e.h b/drivers/net/ethernet/intel/i40e/i40e.h index c8cfe62d5e05..2e02cc68cd3f 100644 --- a/drivers/net/ethernet/intel/i40e/i40e.h +++ b/drivers/net/ethernet/intel/i40e/i40e.h @@ -847,12 +847,12 @@ struct i40e_vsi { struct rtnl_link_stats64 net_stats_offsets; struct i40e_eth_stats eth_stats; struct i40e_eth_stats eth_stats_offsets; - u32 tx_restart; - u32 tx_busy; + u64 tx_restart; + u64 tx_busy; u64 tx_linearize; u64 tx_force_wb; - u32 rx_buf_failed; - u32 rx_page_failed; + u64 rx_buf_failed; + u64 rx_page_failed; /* These are containers of ring pointers, allocated at run-time */ struct i40e_ring **rx_rings; diff --git a/drivers/net/ethernet/intel/i40e/i40e_debugfs.c b/drivers/net/ethernet/intel/i40e/i40e_debugfs.c index 2c1b1da1220e..1e57cc8c47d7 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_debugfs.c +++ b/drivers/net/ethernet/intel/i40e/i40e_debugfs.c @@ -240,7 +240,7 @@ static void i40e_dbg_dump_vsi_seid(struct i40e_pf *pf, int seid) (unsigned long int)vsi->net_stats_offsets.rx_compressed, (unsigned long int)vsi->net_stats_offsets.tx_compressed); dev_info(&pf->pdev->dev, - " tx_restart = %d, tx_busy = %d, rx_buf_failed = %d, rx_page_failed = %d\n", + " tx_restart = %llu, tx_busy = %llu, rx_buf_failed = %llu, rx_page_failed = %llu\n", vsi->tx_restart, vsi->tx_busy, vsi->rx_buf_failed, vsi->rx_page_failed); rcu_read_lock(); diff --git a/drivers/net/ethernet/intel/i40e/i40e_main.c b/drivers/net/ethernet/intel/i40e/i40e_main.c index 9b65cb50282c..f70c478dafdb 100644 --- a/drivers/net/ethernet/intel/i40e/i40e_main.c +++ b/drivers/net/ethernet/intel/i40e/i40e_main.c @@ -778,9 +778,9 @@ static void i40e_update_vsi_stats(struct i40e_vsi *vsi) struct rtnl_link_stats64 *ns; /* netdev stats */ struct i40e_eth_stats *oes; struct i40e_eth_stats *es; /* device's eth stats */ - u32 tx_restart, tx_busy; + u64 tx_restart, tx_busy; struct i40e_ring *p; - u32 rx_page, rx_buf; + u64 rx_page, rx_buf; u64 bytes, packets; unsigned int start; u64 tx_linearize;