diff options
Diffstat (limited to 'drivers/net/wireless/intel/iwlwifi/pcie')
| -rw-r--r-- | drivers/net/wireless/intel/iwlwifi/pcie/ctxt-info-gen3.c | 36 | ||||
| -rw-r--r-- | drivers/net/wireless/intel/iwlwifi/pcie/ctxt-info.c | 4 | ||||
| -rw-r--r-- | drivers/net/wireless/intel/iwlwifi/pcie/drv.c | 159 | ||||
| -rw-r--r-- | drivers/net/wireless/intel/iwlwifi/pcie/internal.h | 295 | ||||
| -rw-r--r-- | drivers/net/wireless/intel/iwlwifi/pcie/rx.c | 20 | ||||
| -rw-r--r-- | drivers/net/wireless/intel/iwlwifi/pcie/trans-gen2.c | 15 | ||||
| -rw-r--r-- | drivers/net/wireless/intel/iwlwifi/pcie/trans.c | 322 | ||||
| -rw-r--r-- | drivers/net/wireless/intel/iwlwifi/pcie/tx-gen2.c | 1188 | ||||
| -rw-r--r-- | drivers/net/wireless/intel/iwlwifi/pcie/tx.c | 1234 |
9 files changed, 2923 insertions, 350 deletions
diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/ctxt-info-gen3.c b/drivers/net/wireless/intel/iwlwifi/pcie/ctxt-info-gen3.c index c8fc8b4fd85c..ae93a72542b2 100644 --- a/drivers/net/wireless/intel/iwlwifi/pcie/ctxt-info-gen3.c +++ b/drivers/net/wireless/intel/iwlwifi/pcie/ctxt-info-gen3.c @@ -1,13 +1,34 @@ // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause /* - * Copyright (C) 2018-2023 Intel Corporation + * Copyright (C) 2018-2024 Intel Corporation */ +#include <linux/dmi.h> #include "iwl-trans.h" #include "iwl-fh.h" #include "iwl-context-info-gen3.h" #include "internal.h" #include "iwl-prph.h" +static const struct dmi_system_id dmi_force_scu_active_approved_list[] = { + { .ident = "DELL", + .matches = { + DMI_MATCH(DMI_SYS_VENDOR, "Dell Inc."), + }, + }, + { .ident = "DELL", + .matches = { + DMI_MATCH(DMI_SYS_VENDOR, "Alienware"), + }, + }, + /* keep last */ + {} +}; + +static bool iwl_is_force_scu_active_approved(void) +{ + return !!dmi_check_system(dmi_force_scu_active_approved_list); +} + static void iwl_pcie_ctxt_info_dbg_enable(struct iwl_trans *trans, struct iwl_prph_scratch_hwm_cfg *dbg_cfg, @@ -68,7 +89,8 @@ iwl_pcie_ctxt_info_dbg_enable(struct iwl_trans *trans, } break; default: - IWL_ERR(trans, "WRT: Invalid buffer destination\n"); + IWL_DEBUG_FW(trans, "WRT: Invalid buffer destination (%d)\n", + le32_to_cpu(fw_mon_cfg->buf_location)); } out: if (dbg_flags) @@ -128,6 +150,14 @@ int iwl_pcie_ctxt_info_gen3_init(struct iwl_trans *trans, if (trans->trans_cfg->imr_enabled) control_flags |= IWL_PRPH_SCRATCH_IMR_DEBUG_EN; + if (CSR_HW_REV_TYPE(trans->hw_rev) == IWL_CFG_MAC_TYPE_GL && + iwl_is_force_scu_active_approved()) { + control_flags |= IWL_PRPH_SCRATCH_SCU_FORCE_ACTIVE; + IWL_DEBUG_FW(trans, + "Context Info: Set SCU_FORCE_ACTIVE (0x%x) in control_flags\n", + IWL_PRPH_SCRATCH_SCU_FORCE_ACTIVE); + } + /* initialize RX default queue */ prph_sc_ctrl->rbd_cfg.free_rbd_addr = cpu_to_le64(trans_pcie->rxq->bd_dma); @@ -187,7 +217,7 @@ int iwl_pcie_ctxt_info_gen3_init(struct iwl_trans *trans, ctxt_info_gen3->cr_tail_idx_arr_base_addr = cpu_to_le64(trans_pcie->prph_info_dma_addr + 3 * PAGE_SIZE / 4); ctxt_info_gen3->mtr_base_addr = - cpu_to_le64(trans->txqs.txq[trans->txqs.cmd.q_id]->dma_addr); + cpu_to_le64(trans_pcie->txqs.txq[trans_pcie->txqs.cmd.q_id]->dma_addr); ctxt_info_gen3->mcr_base_addr = cpu_to_le64(trans_pcie->rxq->used_bd_dma); ctxt_info_gen3->mtr_size = diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/ctxt-info.c b/drivers/net/wireless/intel/iwlwifi/pcie/ctxt-info.c index 0fa92704cd14..344e4d5a1c6e 100644 --- a/drivers/net/wireless/intel/iwlwifi/pcie/ctxt-info.c +++ b/drivers/net/wireless/intel/iwlwifi/pcie/ctxt-info.c @@ -1,7 +1,7 @@ // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause /* * Copyright (C) 2017 Intel Deutschland GmbH - * Copyright (C) 2018-2023 Intel Corporation + * Copyright (C) 2018-2024 Intel Corporation */ #include "iwl-trans.h" #include "iwl-fh.h" @@ -218,7 +218,7 @@ int iwl_pcie_ctxt_info_init(struct iwl_trans *trans, /* initialize TX command queue */ ctxt_info->hcmd_cfg.cmd_queue_addr = - cpu_to_le64(trans->txqs.txq[trans->txqs.cmd.q_id]->dma_addr); + cpu_to_le64(trans_pcie->txqs.txq[trans_pcie->txqs.cmd.q_id]->dma_addr); ctxt_info->hcmd_cfg.cmd_queue_size = TFD_QUEUE_CB_SIZE(IWL_CMD_QUEUE_SIZE); diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/drv.c b/drivers/net/wireless/intel/iwlwifi/pcie/drv.c index 4a657036b9d6..805fb249a0c6 100644 --- a/drivers/net/wireless/intel/iwlwifi/pcie/drv.c +++ b/drivers/net/wireless/intel/iwlwifi/pcie/drv.c @@ -33,7 +33,7 @@ extern int _invalid_type; .driver_data = _ASSIGN_CFG(cfg) /* Hardware specific file defines the PCI IDs table for that hardware module */ -static const struct pci_device_id iwl_hw_card_ids[] = { +VISIBLE_IF_IWLWIFI_KUNIT const struct pci_device_id iwl_hw_card_ids[] = { #if IS_ENABLED(CONFIG_IWLDVM) {IWL_PCI_DEVICE(0x4232, 0x1201, iwl5100_agn_cfg)}, /* Mini Card */ {IWL_PCI_DEVICE(0x4232, 0x1301, iwl5100_agn_cfg)}, /* Half Mini Card */ @@ -492,7 +492,6 @@ static const struct pci_device_id iwl_hw_card_ids[] = { {IWL_PCI_DEVICE(0x7AF0, PCI_ANY_ID, iwl_so_trans_cfg)}, {IWL_PCI_DEVICE(0x51F0, PCI_ANY_ID, iwl_so_long_latency_trans_cfg)}, {IWL_PCI_DEVICE(0x51F1, PCI_ANY_ID, iwl_so_long_latency_imr_trans_cfg)}, - {IWL_PCI_DEVICE(0x51F1, PCI_ANY_ID, iwl_so_long_latency_trans_cfg)}, {IWL_PCI_DEVICE(0x54F0, PCI_ANY_ID, iwl_so_long_latency_trans_cfg)}, {IWL_PCI_DEVICE(0x7F70, PCI_ANY_ID, iwl_so_trans_cfg)}, @@ -501,11 +500,40 @@ static const struct pci_device_id iwl_hw_card_ids[] = { {IWL_PCI_DEVICE(0x7E40, PCI_ANY_ID, iwl_ma_trans_cfg)}, /* Bz devices */ - {IWL_PCI_DEVICE(0x2727, PCI_ANY_ID, iwl_bz_trans_cfg)}, - {IWL_PCI_DEVICE(0x272D, PCI_ANY_ID, iwl_bz_trans_cfg)}, - {IWL_PCI_DEVICE(0x272b, PCI_ANY_ID, iwl_bz_trans_cfg)}, - {IWL_PCI_DEVICE(0xA840, PCI_ANY_ID, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0x272b, PCI_ANY_ID, iwl_gl_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x0000, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x0090, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x0094, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x0098, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x009C, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x00C0, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x00C4, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x00E0, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x00E4, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x00E8, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x00EC, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x0100, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x0110, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x0114, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x0118, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x011C, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x0310, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x0314, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x0510, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x0A10, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x1671, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x1672, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x1771, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x1772, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x1791, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x1792, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x4090, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x40C4, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x40E0, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x4110, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0xA840, 0x4314, iwl_bz_trans_cfg)}, {IWL_PCI_DEVICE(0x7740, PCI_ANY_ID, iwl_bz_trans_cfg)}, + {IWL_PCI_DEVICE(0x4D40, PCI_ANY_ID, iwl_bz_trans_cfg)}, /* Sc devices */ {IWL_PCI_DEVICE(0xE440, PCI_ANY_ID, iwl_sc_trans_cfg)}, @@ -517,6 +545,7 @@ static const struct pci_device_id iwl_hw_card_ids[] = { {0} }; MODULE_DEVICE_TABLE(pci, iwl_hw_card_ids); +EXPORT_SYMBOL_IF_IWLWIFI_KUNIT(iwl_hw_card_ids); #define _IWL_DEV_INFO(_device, _subdevice, _mac_type, _mac_step, _rf_type, \ _rf_id, _rf_step, _no_160, _cores, _cdb, _cfg, _name) \ @@ -946,11 +975,6 @@ VISIBLE_IF_IWLWIFI_KUNIT const struct iwl_dev_info iwl_dev_info_table[] = { iwl_cfg_ma, iwl_ax211_name), _IWL_DEV_INFO(IWL_CFG_ANY, IWL_CFG_ANY, IWL_CFG_MAC_TYPE_MA, IWL_CFG_ANY, - IWL_CFG_RF_TYPE_MR, IWL_CFG_ANY, IWL_CFG_ANY, - IWL_CFG_ANY, IWL_CFG_ANY, IWL_CFG_NO_CDB, - iwl_cfg_ma, iwl_ax221_name), - _IWL_DEV_INFO(IWL_CFG_ANY, IWL_CFG_ANY, - IWL_CFG_MAC_TYPE_MA, IWL_CFG_ANY, IWL_CFG_RF_TYPE_FM, IWL_CFG_ANY, IWL_CFG_ANY, IWL_CFG_ANY, IWL_CFG_ANY, IWL_CFG_NO_CDB, iwl_cfg_ma, iwl_ax231_name), @@ -1001,25 +1025,6 @@ VISIBLE_IF_IWLWIFI_KUNIT const struct iwl_dev_info iwl_dev_info_table[] = { IWL_CFG_160, IWL_CFG_ANY, IWL_CFG_CDB, iwlax411_2ax_cfg_so_gf4_a0, iwl_ax411_name), -/* Bz */ - _IWL_DEV_INFO(IWL_CFG_ANY, IWL_CFG_ANY, - IWL_CFG_MAC_TYPE_BZ, IWL_CFG_ANY, - IWL_CFG_ANY, IWL_CFG_ANY, IWL_CFG_ANY, - IWL_CFG_ANY, IWL_CFG_ANY, IWL_CFG_ANY, - iwl_cfg_bz, iwl_bz_name), - -/* Ga (Gl) */ - _IWL_DEV_INFO(IWL_CFG_ANY, IWL_CFG_ANY, - IWL_CFG_MAC_TYPE_GL, IWL_CFG_ANY, - IWL_CFG_RF_TYPE_FM, IWL_CFG_ANY, IWL_CFG_ANY, - IWL_CFG_320, IWL_CFG_ANY, IWL_CFG_NO_CDB, - iwl_cfg_gl, iwl_bz_name), - _IWL_DEV_INFO(IWL_CFG_ANY, IWL_CFG_ANY, - IWL_CFG_MAC_TYPE_GL, IWL_CFG_ANY, - IWL_CFG_RF_TYPE_FM, IWL_CFG_ANY, IWL_CFG_ANY, - IWL_CFG_NO_320, IWL_CFG_ANY, IWL_CFG_NO_CDB, - iwl_cfg_gl, iwl_mtp_name), - /* SoF with JF2 */ _IWL_DEV_INFO(IWL_CFG_ANY, IWL_CFG_ANY, IWL_CFG_MAC_TYPE_SOF, IWL_CFG_ANY, @@ -1100,23 +1105,31 @@ VISIBLE_IF_IWLWIFI_KUNIT const struct iwl_dev_info iwl_dev_info_table[] = { IWL_CFG_NO_160, IWL_CFG_CORES_BT, IWL_CFG_NO_CDB, iwlax210_2ax_cfg_so_jf_b0, iwl9462_name), -/* MsP */ -/* For now we use the same FW as MR, but this will change in the future. */ +/* Bz */ +/* FIXME: need to change the naming according to the actual CRF */ _IWL_DEV_INFO(IWL_CFG_ANY, IWL_CFG_ANY, - IWL_CFG_MAC_TYPE_SO, IWL_CFG_ANY, - IWL_CFG_RF_TYPE_MS, IWL_CFG_ANY, IWL_CFG_ANY, - IWL_CFG_160, IWL_CFG_ANY, IWL_CFG_NO_CDB, - iwl_cfg_so_a0_ms_a0, iwl_ax204_name), + IWL_CFG_MAC_TYPE_BZ, IWL_CFG_ANY, + IWL_CFG_ANY, IWL_CFG_ANY, IWL_CFG_ANY, + IWL_CFG_ANY, IWL_CFG_ANY, IWL_CFG_ANY, + iwl_cfg_bz, iwl_fm_name), + _IWL_DEV_INFO(IWL_CFG_ANY, IWL_CFG_ANY, - IWL_CFG_MAC_TYPE_SOF, IWL_CFG_ANY, - IWL_CFG_RF_TYPE_MS, IWL_CFG_ANY, IWL_CFG_ANY, - IWL_CFG_160, IWL_CFG_ANY, IWL_CFG_NO_CDB, - iwl_cfg_so_a0_ms_a0, iwl_ax204_name), + IWL_CFG_MAC_TYPE_BZ_W, IWL_CFG_ANY, + IWL_CFG_ANY, IWL_CFG_ANY, IWL_CFG_ANY, + IWL_CFG_ANY, IWL_CFG_ANY, IWL_CFG_ANY, + iwl_cfg_bz, iwl_fm_name), + +/* Ga (Gl) */ _IWL_DEV_INFO(IWL_CFG_ANY, IWL_CFG_ANY, - IWL_CFG_MAC_TYPE_MA, IWL_CFG_ANY, - IWL_CFG_RF_TYPE_MS, IWL_CFG_ANY, IWL_CFG_ANY, - IWL_CFG_160, IWL_CFG_ANY, IWL_CFG_NO_CDB, - iwl_cfg_ma, iwl_ax204_name), + IWL_CFG_MAC_TYPE_GL, IWL_CFG_ANY, + IWL_CFG_RF_TYPE_FM, IWL_CFG_ANY, IWL_CFG_ANY, + IWL_CFG_320, IWL_CFG_ANY, IWL_CFG_NO_CDB, + iwl_cfg_gl, iwl_gl_name), + _IWL_DEV_INFO(IWL_CFG_ANY, IWL_CFG_ANY, + IWL_CFG_MAC_TYPE_GL, IWL_CFG_ANY, + IWL_CFG_RF_TYPE_FM, IWL_CFG_ANY, IWL_CFG_ANY, + IWL_CFG_NO_320, IWL_CFG_ANY, IWL_CFG_NO_CDB, + iwl_cfg_gl, iwl_mtp_name), /* Sc */ _IWL_DEV_INFO(IWL_CFG_ANY, IWL_CFG_ANY, @@ -1150,6 +1163,7 @@ static void get_crf_id(struct iwl_trans *iwl_trans) { u32 sd_reg_ver_addr; u32 val = 0; + u8 step; if (iwl_trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_AX210) sd_reg_ver_addr = SD_REG_VER_GEN2; @@ -1168,16 +1182,23 @@ static void get_crf_id(struct iwl_trans *iwl_trans) iwl_trans->hw_cnv_id = iwl_read_prph_no_grab(iwl_trans, CNVI_AUX_MISC_CHIP); + /* For BZ-W, take B step also when A step is indicated */ + if (CSR_HW_REV_TYPE(iwl_trans->hw_rev) == IWL_CFG_MAC_TYPE_BZ_W) + step = SILICON_B_STEP; + /* In BZ, the MAC step must be read from the CNVI aux register */ if (CSR_HW_REV_TYPE(iwl_trans->hw_rev) == IWL_CFG_MAC_TYPE_BZ) { - u8 step = CNVI_AUX_MISC_CHIP_MAC_STEP(iwl_trans->hw_cnv_id); + step = CNVI_AUX_MISC_CHIP_MAC_STEP(iwl_trans->hw_cnv_id); /* For BZ-U, take B step also when A step is indicated */ if ((CNVI_AUX_MISC_CHIP_PROD_TYPE(iwl_trans->hw_cnv_id) == CNVI_AUX_MISC_CHIP_PROD_TYPE_BZ_U) && step == SILICON_A_STEP) step = SILICON_B_STEP; + } + if (CSR_HW_REV_TYPE(iwl_trans->hw_rev) == IWL_CFG_MAC_TYPE_BZ || + CSR_HW_REV_TYPE(iwl_trans->hw_rev) == IWL_CFG_MAC_TYPE_BZ_W) { iwl_trans->hw_rev_step = step; iwl_trans->hw_rev |= step; } @@ -1224,12 +1245,7 @@ static int map_crf_id(struct iwl_trans *iwl_trans) case REG_CRF_ID_TYPE_GF: iwl_trans->hw_rf_id = (IWL_CFG_RF_TYPE_GF << 12); break; - case REG_CRF_ID_TYPE_MR: - iwl_trans->hw_rf_id = (IWL_CFG_RF_TYPE_MR << 12); - break; case REG_CRF_ID_TYPE_FM: - case REG_CRF_ID_TYPE_FMI: - case REG_CRF_ID_TYPE_FMR: iwl_trans->hw_rf_id = (IWL_CFG_RF_TYPE_FM << 12); break; case REG_CRF_ID_TYPE_WHP: @@ -1488,6 +1504,8 @@ static int iwl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *ent) if (!iwl_trans->name) iwl_trans->name = iwl_trans->cfg->name; + IWL_INFO(iwl_trans, "Detected %s\n", iwl_trans->name); + if (iwl_trans->trans_cfg->mq_rx_supported) { if (WARN_ON(!iwl_trans->cfg->num_rbds)) { ret = -EINVAL; @@ -1557,11 +1575,12 @@ static int iwl_pci_suspend(struct device *device) return 0; } -static int iwl_pci_resume(struct device *device) +static int _iwl_pci_resume(struct device *device, bool restore) { struct pci_dev *pdev = to_pci_dev(device); struct iwl_trans *trans = pci_get_drvdata(pdev); struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + bool device_was_powered_off = false; /* Before you put code here, think about WoWLAN. You cannot check here * whether WoWLAN is enabled or not, and your code will run even if @@ -1577,6 +1596,26 @@ static int iwl_pci_resume(struct device *device) if (!trans->op_mode) return 0; + /* + * Scratch value was altered, this means the device was powered off, we + * need to reset it completely. + * Note: MAC (bits 0:7) will be cleared upon suspend even with wowlan, + * so assume that any bits there mean that the device is usable. + */ + if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_BZ && + !iwl_read32(trans, CSR_FUNC_SCRATCH)) + device_was_powered_off = true; + + if (restore || device_was_powered_off) { + trans->state = IWL_TRANS_NO_FW; + /* Hope for the best here ... If one of those steps fails we + * won't really know how to recover. + */ + iwl_pcie_prepare_card_hw(trans); + iwl_finish_nic_init(trans); + iwl_op_mode_device_powered_off(trans->op_mode); + } + /* In WOWLAN, let iwl_trans_pcie_d3_resume do the rest of the work */ if (test_bit(STATUS_DEVICE_ENABLED, &trans->status)) return 0; @@ -1597,9 +1636,23 @@ static int iwl_pci_resume(struct device *device) return 0; } +static int iwl_pci_restore(struct device *device) +{ + return _iwl_pci_resume(device, true); +} + +static int iwl_pci_resume(struct device *device) +{ + return _iwl_pci_resume(device, false); +} + static const struct dev_pm_ops iwl_dev_pm_ops = { - SET_SYSTEM_SLEEP_PM_OPS(iwl_pci_suspend, - iwl_pci_resume) + .suspend = pm_sleep_ptr(iwl_pci_suspend), + .resume = pm_sleep_ptr(iwl_pci_resume), + .freeze = pm_sleep_ptr(iwl_pci_suspend), + .thaw = pm_sleep_ptr(iwl_pci_resume), + .poweroff = pm_sleep_ptr(iwl_pci_suspend), + .restore = pm_sleep_ptr(iwl_pci_restore), }; #define IWL_PM_OPS (&iwl_dev_pm_ops) diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/internal.h b/drivers/net/wireless/intel/iwlwifi/pcie/internal.h index 7805a42948af..27a7e0b5b3d5 100644 --- a/drivers/net/wireless/intel/iwlwifi/pcie/internal.h +++ b/drivers/net/wireless/intel/iwlwifi/pcie/internal.h @@ -1,6 +1,6 @@ /* SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause */ /* - * Copyright (C) 2003-2015, 2018-2023 Intel Corporation + * Copyright (C) 2003-2015, 2018-2024 Intel Corporation * Copyright (C) 2013-2015 Intel Mobile Communications GmbH * Copyright (C) 2016-2017 Intel Deutschland GmbH */ @@ -22,7 +22,6 @@ #include "iwl-io.h" #include "iwl-op-mode.h" #include "iwl-drv.h" -#include "queue/tx.h" #include "iwl-context-info.h" /* @@ -273,7 +272,7 @@ enum iwl_pcie_fw_reset_state { }; /** - * enum wl_pcie_imr_status - imr dma transfer state + * enum iwl_pcie_imr_status - imr dma transfer state * @IMR_D2S_IDLE: default value of the dma transfer * @IMR_D2S_REQUESTED: dma transfer requested * @IMR_D2S_COMPLETED: dma transfer completed @@ -287,6 +286,58 @@ enum iwl_pcie_imr_status { }; /** + * struct iwl_pcie_txqs - TX queues data + * + * @bc_table_dword: true if the BC table expects DWORD (as opposed to bytes) + * @page_offs: offset from skb->cb to mac header page pointer + * @dev_cmd_offs: offset from skb->cb to iwl_device_tx_cmd pointer + * @queue_used: bit mask of used queues + * @queue_stopped: bit mask of stopped queues + * @txq: array of TXQ data structures representing the TXQs + * @scd_bc_tbls: gen1 pointer to the byte count table of the scheduler + * @queue_alloc_cmd_ver: queue allocation command version + * @bc_pool: bytecount DMA allocations pool + * @bc_tbl_size: bytecount table size + * @tso_hdr_page: page allocated (per CPU) for A-MSDU headers when doing TSO + * (and similar usage) + * @cmd: command queue data + * @cmd.fifo: FIFO number + * @cmd.q_id: queue ID + * @cmd.wdg_timeout: watchdog timeout + * @tfd: TFD data + * @tfd.max_tbs: max number of buffers per TFD + * @tfd.size: TFD size + * @tfd.addr_size: TFD/TB address size + */ +struct iwl_pcie_txqs { + unsigned long queue_used[BITS_TO_LONGS(IWL_MAX_TVQM_QUEUES)]; + unsigned long queue_stopped[BITS_TO_LONGS(IWL_MAX_TVQM_QUEUES)]; + struct iwl_txq *txq[IWL_MAX_TVQM_QUEUES]; + struct dma_pool *bc_pool; + size_t bc_tbl_size; + bool bc_table_dword; + u8 page_offs; + u8 dev_cmd_offs; + struct iwl_tso_hdr_page __percpu *tso_hdr_page; + + struct { + u8 fifo; + u8 q_id; + unsigned int wdg_timeout; + } cmd; + + struct { + u8 max_tbs; + u16 size; + u8 addr_size; + } tfd; + + struct iwl_dma_ptr scd_bc_tbls; + + u8 queue_alloc_cmd_ver; +}; + +/** * struct iwl_trans_pcie - PCIe transport specific data * @rxq: all the RX queue data * @rx_pool: initial pool of iwl_rx_mem_buffer for all the queues @@ -367,6 +418,7 @@ enum iwl_pcie_imr_status { * @is_down: indicates the NIC is down * @isr_stats: interrupt statistics * @napi_dev: (fake) netdev for NAPI registration + * @txqs: transport tx queues data. */ struct iwl_trans_pcie { struct iwl_rxq *rxq; @@ -386,7 +438,7 @@ struct iwl_trans_pcie { dma_addr_t iml_dma_addr; struct iwl_trans *trans; - struct net_device napi_dev; + struct net_device *napi_dev; /* INT ICT Table */ __le32 *ict_tbl; @@ -464,6 +516,8 @@ struct iwl_trans_pcie { enum iwl_pcie_imr_status imr_status; wait_queue_head_t imr_waitq; char rf_name[32]; + + struct iwl_pcie_txqs txqs; }; static inline struct iwl_trans_pcie * @@ -538,6 +592,33 @@ void iwl_pcie_disable_ict(struct iwl_trans *trans); /***************************************************** * TX / HCMD ******************************************************/ +/* We need 2 entries for the TX command and header, and another one might + * be needed for potential data in the SKB's head. The remaining ones can + * be used for frags. + */ +#define IWL_TRANS_PCIE_MAX_FRAGS(trans_pcie) ((trans_pcie)->txqs.tfd.max_tbs - 3) + +struct iwl_tso_hdr_page { + struct page *page; + u8 *pos; +}; + +/* + * Note that we put this struct *last* in the page. By doing that, we ensure + * that no TB referencing this page can trigger the 32-bit boundary hardware + * bug. + */ +struct iwl_tso_page_info { + dma_addr_t dma_addr; + struct page *next; + refcount_t use_count; +}; + +#define IWL_TSO_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(struct iwl_tso_page_info)) +#define IWL_TSO_PAGE_INFO(addr) \ + ((struct iwl_tso_page_info *)(((unsigned long)addr & PAGE_MASK) + \ + IWL_TSO_PAGE_DATA_SIZE)) + int iwl_pcie_tx_init(struct iwl_trans *trans); void iwl_pcie_tx_start(struct iwl_trans *trans, u32 scd_base_addr); int iwl_pcie_tx_stop(struct iwl_trans *trans); @@ -552,10 +633,171 @@ void iwl_trans_pcie_txq_set_shared_mode(struct iwl_trans *trans, u32 txq_id, int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb, struct iwl_device_tx_cmd *dev_cmd, int txq_id); void iwl_pcie_txq_check_wrptrs(struct iwl_trans *trans); -int iwl_trans_pcie_send_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd); void iwl_pcie_hcmd_complete(struct iwl_trans *trans, struct iwl_rx_cmd_buffer *rxb); void iwl_trans_pcie_tx_reset(struct iwl_trans *trans); +int iwl_pcie_txq_alloc(struct iwl_trans *trans, struct iwl_txq *txq, + int slots_num, bool cmd_queue); + +dma_addr_t iwl_pcie_get_sgt_tb_phys(struct sg_table *sgt, unsigned int offset, + unsigned int len); +struct sg_table *iwl_pcie_prep_tso(struct iwl_trans *trans, struct sk_buff *skb, + struct iwl_cmd_meta *cmd_meta, + u8 **hdr, unsigned int hdr_room); + +void iwl_pcie_free_tso_pages(struct iwl_trans *trans, struct sk_buff *skb, + struct iwl_cmd_meta *cmd_meta); + +static inline dma_addr_t iwl_pcie_get_tso_page_phys(void *addr) +{ + dma_addr_t res; + + res = IWL_TSO_PAGE_INFO(addr)->dma_addr; + res += (unsigned long)addr & ~PAGE_MASK; + + return res; +} + +static inline dma_addr_t +iwl_txq_get_first_tb_dma(struct iwl_txq *txq, int idx) +{ + return txq->first_tb_dma + + sizeof(struct iwl_pcie_first_tb_buf) * idx; +} + +static inline u16 iwl_txq_get_cmd_index(const struct iwl_txq *q, u32 index) +{ + return index & (q->n_window - 1); +} + +static inline void *iwl_txq_get_tfd(struct iwl_trans *trans, + struct iwl_txq *txq, int idx) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + + if (trans->trans_cfg->gen2) + idx = iwl_txq_get_cmd_index(txq, idx); + + return (u8 *)txq->tfds + trans_pcie->txqs.tfd.size * idx; +} + +/* + * We need this inline in case dma_addr_t is only 32-bits - since the + * hardware is always 64-bit, the issue can still occur in that case, + * so use u64 for 'phys' here to force the addition in 64-bit. + */ +static inline bool iwl_txq_crosses_4g_boundary(u64 phys, u16 len) +{ + return upper_32_bits(phys) != upper_32_bits(phys + len); +} + +int iwl_txq_space(struct iwl_trans *trans, const struct iwl_txq *q); + +static inline void iwl_txq_stop(struct iwl_trans *trans, struct iwl_txq *txq) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + + if (!test_and_set_bit(txq->id, trans_pcie->txqs.queue_stopped)) { + iwl_op_mode_queue_full(trans->op_mode, txq->id); + IWL_DEBUG_TX_QUEUES(trans, "Stop hwq %d\n", txq->id); + } else { + IWL_DEBUG_TX_QUEUES(trans, "hwq %d already stopped\n", + txq->id); + } +} + +/** + * iwl_txq_inc_wrap - increment queue index, wrap back to beginning + * @trans: the transport (for configuration data) + * @index: current index + */ +static inline int iwl_txq_inc_wrap(struct iwl_trans *trans, int index) +{ + return ++index & + (trans->trans_cfg->base_params->max_tfd_queue_size - 1); +} + +/** + * iwl_txq_dec_wrap - decrement queue index, wrap back to end + * @trans: the transport (for configuration data) + * @index: current index + */ +static inline int iwl_txq_dec_wrap(struct iwl_trans *trans, int index) +{ + return --index & + (trans->trans_cfg->base_params->max_tfd_queue_size - 1); +} + +void iwl_txq_log_scd_error(struct iwl_trans *trans, struct iwl_txq *txq); + +static inline void +iwl_trans_pcie_wake_queue(struct iwl_trans *trans, struct iwl_txq *txq) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + + if (test_and_clear_bit(txq->id, trans_pcie->txqs.queue_stopped)) { + IWL_DEBUG_TX_QUEUES(trans, "Wake hwq %d\n", txq->id); + iwl_op_mode_queue_not_full(trans->op_mode, txq->id); + } +} + +int iwl_txq_gen2_set_tb(struct iwl_trans *trans, + struct iwl_tfh_tfd *tfd, dma_addr_t addr, + u16 len); + +static inline void iwl_txq_set_tfd_invalid_gen2(struct iwl_trans *trans, + struct iwl_tfh_tfd *tfd) +{ + tfd->num_tbs = 0; + + iwl_txq_gen2_set_tb(trans, tfd, trans->invalid_tx_cmd.dma, + trans->invalid_tx_cmd.size); +} + +void iwl_txq_gen2_tfd_unmap(struct iwl_trans *trans, + struct iwl_cmd_meta *meta, + struct iwl_tfh_tfd *tfd); + +int iwl_txq_dyn_alloc(struct iwl_trans *trans, u32 flags, + u32 sta_mask, u8 tid, + int size, unsigned int timeout); + +int iwl_txq_gen2_tx(struct iwl_trans *trans, struct sk_buff *skb, + struct iwl_device_tx_cmd *dev_cmd, int txq_id); + +void iwl_txq_dyn_free(struct iwl_trans *trans, int queue); +void iwl_txq_gen2_tx_free(struct iwl_trans *trans); +int iwl_txq_init(struct iwl_trans *trans, struct iwl_txq *txq, + int slots_num, bool cmd_queue); +int iwl_txq_gen2_init(struct iwl_trans *trans, int txq_id, + int queue_size); + +static inline u16 iwl_txq_gen1_tfd_tb_get_len(struct iwl_trans *trans, + void *_tfd, u8 idx) +{ + struct iwl_tfd *tfd; + struct iwl_tfd_tb *tb; + + if (trans->trans_cfg->gen2) { + struct iwl_tfh_tfd *tfh_tfd = _tfd; + struct iwl_tfh_tb *tfh_tb = &tfh_tfd->tbs[idx]; + + return le16_to_cpu(tfh_tb->tb_len); + } + + tfd = (struct iwl_tfd *)_tfd; + tb = &tfd->tbs[idx]; + + return le16_to_cpu(tb->hi_n_len) >> 4; +} + +void iwl_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn, + struct sk_buff_head *skbs, bool is_flush); +void iwl_pcie_set_q_ptrs(struct iwl_trans *trans, int txq_id, int ptr); +void iwl_pcie_freeze_txq_timer(struct iwl_trans *trans, + unsigned long txqs, bool freeze); +int iwl_trans_pcie_wait_txq_empty(struct iwl_trans *trans, int txq_idx); +int iwl_trans_pcie_wait_txqs_empty(struct iwl_trans *trans, u32 txq_bm); /***************************************************** * Error handling @@ -822,12 +1064,51 @@ void iwl_trans_pcie_dump_regs(struct iwl_trans *trans); #ifdef CONFIG_IWLWIFI_DEBUGFS void iwl_trans_pcie_dbgfs_register(struct iwl_trans *trans); +void iwl_trans_pcie_debugfs_cleanup(struct iwl_trans *trans); #else static inline void iwl_trans_pcie_dbgfs_register(struct iwl_trans *trans) { } #endif void iwl_pcie_rx_allocator_work(struct work_struct *data); +/* common trans ops for all generations transports */ +void iwl_trans_pcie_configure(struct iwl_trans *trans, + const struct iwl_trans_config *trans_cfg); +int iwl_trans_pcie_start_hw(struct iwl_trans *trans); +void iwl_trans_pcie_op_mode_leave(struct iwl_trans *trans); +void iwl_trans_pcie_write8(struct iwl_trans *trans, u32 ofs, u8 val); +void iwl_trans_pcie_write32(struct iwl_trans *trans, u32 ofs, u32 val); +u32 iwl_trans_pcie_read32(struct iwl_trans *trans, u32 ofs); +u32 iwl_trans_pcie_read_prph(struct iwl_trans *trans, u32 reg); +void iwl_trans_pcie_write_prph(struct iwl_trans *trans, u32 addr, u32 val); +int iwl_trans_pcie_read_mem(struct iwl_trans *trans, u32 addr, + void *buf, int dwords); +int iwl_trans_pcie_write_mem(struct iwl_trans *trans, u32 addr, + const void *buf, int dwords); +int iwl_trans_pcie_sw_reset(struct iwl_trans *trans, bool retake_ownership); +struct iwl_trans_dump_data * +iwl_trans_pcie_dump_data(struct iwl_trans *trans, u32 dump_mask, + const struct iwl_dump_sanitize_ops *sanitize_ops, + void *sanitize_ctx); +int iwl_trans_pcie_d3_resume(struct iwl_trans *trans, + enum iwl_d3_status *status, + bool test, bool reset); +int iwl_trans_pcie_d3_suspend(struct iwl_trans *trans, bool test, bool reset); +void iwl_trans_pci_interrupts(struct iwl_trans *trans, bool enable); +void iwl_trans_pcie_sync_nmi(struct iwl_trans *trans); +void iwl_trans_pcie_set_bits_mask(struct iwl_trans *trans, u32 reg, + u32 mask, u32 value); +int iwl_trans_pcie_read_config32(struct iwl_trans *trans, u32 ofs, + u32 *val); +bool iwl_trans_pcie_grab_nic_access(struct iwl_trans *trans); +void iwl_trans_pcie_release_nic_access(struct iwl_trans *trans); + +/* transport gen 1 exported functions */ +void iwl_trans_pcie_fw_alive(struct iwl_trans *trans, u32 scd_addr); +int iwl_trans_pcie_start_fw(struct iwl_trans *trans, + const struct fw_img *fw, bool run_in_rfkill); +void iwl_trans_pcie_stop_device(struct iwl_trans *trans); + /* common functions that are used by gen2 transport */ int iwl_pcie_gen2_apm_init(struct iwl_trans *trans); void iwl_pcie_apm_config(struct iwl_trans *trans); @@ -849,7 +1130,7 @@ void iwl_pcie_alloc_fw_monitor(struct iwl_trans *trans, u8 max_power); /* transport gen 2 exported functions */ int iwl_trans_pcie_gen2_start_fw(struct iwl_trans *trans, const struct fw_img *fw, bool run_in_rfkill); -void iwl_trans_pcie_gen2_fw_alive(struct iwl_trans *trans, u32 scd_addr); +void iwl_trans_pcie_gen2_fw_alive(struct iwl_trans *trans); int iwl_trans_pcie_gen2_send_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd); void iwl_trans_pcie_gen2_stop_device(struct iwl_trans *trans); @@ -864,5 +1145,7 @@ void iwl_trans_pcie_copy_imr_fh(struct iwl_trans *trans, u32 dst_addr, u64 src_addr, u32 byte_cnt); int iwl_trans_pcie_copy_imr(struct iwl_trans *trans, u32 dst_addr, u64 src_addr, u32 byte_cnt); +int iwl_trans_pcie_rxq_dma_data(struct iwl_trans *trans, int queue, + struct iwl_trans_rxq_dma_data *data); #endif /* __iwl_trans_int_pcie_h__ */ diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/rx.c b/drivers/net/wireless/intel/iwlwifi/pcie/rx.c index 9c2461ba13c5..afb88eab8174 100644 --- a/drivers/net/wireless/intel/iwlwifi/pcie/rx.c +++ b/drivers/net/wireless/intel/iwlwifi/pcie/rx.c @@ -1,6 +1,6 @@ // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause /* - * Copyright (C) 2003-2014, 2018-2023 Intel Corporation + * Copyright (C) 2003-2014, 2018-2024 Intel Corporation * Copyright (C) 2013-2015 Intel Mobile Communications GmbH * Copyright (C) 2016-2017 Intel Deutschland GmbH */ @@ -1000,6 +1000,11 @@ void iwl_pcie_rx_init_rxb_lists(struct iwl_rxq *rxq) static int iwl_pcie_rx_handle(struct iwl_trans *trans, int queue, int budget); +static inline struct iwl_trans_pcie *iwl_netdev_to_trans_pcie(struct net_device *dev) +{ + return *(struct iwl_trans_pcie **)netdev_priv(dev); +} + static int iwl_pcie_napi_poll(struct napi_struct *napi, int budget) { struct iwl_rxq *rxq = container_of(napi, struct iwl_rxq, napi); @@ -1007,7 +1012,7 @@ static int iwl_pcie_napi_poll(struct napi_struct *napi, int budget) struct iwl_trans *trans; int ret; - trans_pcie = container_of(napi->dev, struct iwl_trans_pcie, napi_dev); + trans_pcie = iwl_netdev_to_trans_pcie(napi->dev); trans = trans_pcie->trans; ret = iwl_pcie_rx_handle(trans, rxq->id, budget); @@ -1034,7 +1039,7 @@ static int iwl_pcie_napi_poll_msix(struct napi_struct *napi, int budget) struct iwl_trans *trans; int ret; - trans_pcie = container_of(napi->dev, struct iwl_trans_pcie, napi_dev); + trans_pcie = iwl_netdev_to_trans_pcie(napi->dev); trans = trans_pcie->trans; ret = iwl_pcie_rx_handle(trans, rxq->id, budget); @@ -1131,7 +1136,7 @@ static int _iwl_pcie_rx_init(struct iwl_trans *trans) if (trans_pcie->msix_enabled) poll = iwl_pcie_napi_poll_msix; - netif_napi_add(&trans_pcie->napi_dev, &rxq->napi, + netif_napi_add(trans_pcie->napi_dev, &rxq->napi, poll); napi_enable(&rxq->napi); } @@ -1296,7 +1301,7 @@ static void iwl_pcie_rx_handle_rb(struct iwl_trans *trans, int i) { struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); - struct iwl_txq *txq = trans->txqs.txq[trans->txqs.cmd.q_id]; + struct iwl_txq *txq = trans_pcie->txqs.txq[trans_pcie->txqs.cmd.q_id]; bool page_stolen = false; int max_len = trans_pcie->rx_buf_bytes; u32 offset = 0; @@ -1673,6 +1678,7 @@ irqreturn_t iwl_pcie_irq_rx_msix_handler(int irq, void *dev_id) */ static void iwl_pcie_irq_handle_error(struct iwl_trans *trans) { + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); int i; /* W/A for WiFi/WiMAX coex and WiMAX own the RF */ @@ -1689,9 +1695,9 @@ static void iwl_pcie_irq_handle_error(struct iwl_trans *trans) } for (i = 0; i < trans->trans_cfg->base_params->num_of_queues; i++) { - if (!trans->txqs.txq[i]) + if (!trans_pcie->txqs.txq[i]) continue; - del_timer(&trans->txqs.txq[i]->stuck_timer); + del_timer(&trans_pcie->txqs.txq[i]->stuck_timer); } /* The STATUS_FW_ERROR bit is set in this function. This must happen diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/trans-gen2.c b/drivers/net/wireless/intel/iwlwifi/pcie/trans-gen2.c index a4a4772330cf..8903a5692dfb 100644 --- a/drivers/net/wireless/intel/iwlwifi/pcie/trans-gen2.c +++ b/drivers/net/wireless/intel/iwlwifi/pcie/trans-gen2.c @@ -1,7 +1,7 @@ // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause /* * Copyright (C) 2017 Intel Deutschland GmbH - * Copyright (C) 2018-2023 Intel Corporation + * Copyright (C) 2018-2024 Intel Corporation */ #include "iwl-trans.h" #include "iwl-prph.h" @@ -247,7 +247,7 @@ static int iwl_pcie_gen2_nic_init(struct iwl_trans *trans) return -ENOMEM; /* Allocate or reset and init all Tx and Command queues */ - if (iwl_txq_gen2_init(trans, trans->txqs.cmd.q_id, queue_size)) + if (iwl_txq_gen2_init(trans, trans_pcie->txqs.cmd.q_id, queue_size)) return -ENOMEM; /* enable shadow regs in HW */ @@ -339,16 +339,17 @@ static void iwl_pcie_get_rf_name(struct iwl_trans *trans) pos += scnprintf(buf + pos, buflen - pos, "\n"); } -void iwl_trans_pcie_gen2_fw_alive(struct iwl_trans *trans, u32 scd_addr) +void iwl_trans_pcie_gen2_fw_alive(struct iwl_trans *trans) { struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); iwl_pcie_reset_ict(trans); /* make sure all queue are not stopped/used */ - memset(trans->txqs.queue_stopped, 0, - sizeof(trans->txqs.queue_stopped)); - memset(trans->txqs.queue_used, 0, sizeof(trans->txqs.queue_used)); + memset(trans_pcie->txqs.queue_stopped, 0, + sizeof(trans_pcie->txqs.queue_stopped)); + memset(trans_pcie->txqs.queue_used, 0, + sizeof(trans_pcie->txqs.queue_used)); /* now that we got alive we can free the fw image & the context info. * paging memory cannot be freed included since FW will still use it @@ -525,6 +526,8 @@ int iwl_trans_pcie_gen2_start_fw(struct iwl_trans *trans, keep_ram_busy = !iwl_pcie_set_ltr(trans); if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_BZ) { + IWL_DEBUG_POWER(trans, "function scratch register value is 0x%08x\n", + iwl_read32(trans, CSR_FUNC_SCRATCH)); iwl_write32(trans, CSR_FUNC_SCRATCH, CSR_FUNC_SCRATCH_INIT_VALUE); iwl_set_bit(trans, CSR_GP_CNTRL, CSR_GP_CNTRL_REG_FLAG_ROM_START); diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/trans.c b/drivers/net/wireless/intel/iwlwifi/pcie/trans.c index 6c76b2dd6878..3b9943eb6934 100644 --- a/drivers/net/wireless/intel/iwlwifi/pcie/trans.c +++ b/drivers/net/wireless/intel/iwlwifi/pcie/trans.c @@ -1,6 +1,6 @@ // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause /* - * Copyright (C) 2007-2015, 2018-2023 Intel Corporation + * Copyright (C) 2007-2015, 2018-2024 Intel Corporation * Copyright (C) 2013-2015 Intel Mobile Communications GmbH * Copyright (C) 2016-2017 Intel Deutschland GmbH */ @@ -127,8 +127,7 @@ out: kfree(buf); } -static int iwl_trans_pcie_sw_reset(struct iwl_trans *trans, - bool retake_ownership) +int iwl_trans_pcie_sw_reset(struct iwl_trans *trans, bool retake_ownership) { /* Reset entire device - do controller reset (results in SHRD_HW_RST) */ if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_BZ) { @@ -1336,8 +1335,8 @@ void iwl_pcie_synchronize_irqs(struct iwl_trans *trans) } } -static int iwl_trans_pcie_start_fw(struct iwl_trans *trans, - const struct fw_img *fw, bool run_in_rfkill) +int iwl_trans_pcie_start_fw(struct iwl_trans *trans, + const struct fw_img *fw, bool run_in_rfkill) { struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); bool hw_rfkill; @@ -1423,7 +1422,7 @@ out: return ret; } -static void iwl_trans_pcie_fw_alive(struct iwl_trans *trans, u32 scd_addr) +void iwl_trans_pcie_fw_alive(struct iwl_trans *trans, u32 scd_addr) { iwl_pcie_reset_ict(trans); iwl_pcie_tx_start(trans, scd_addr); @@ -1458,7 +1457,7 @@ void iwl_trans_pcie_handle_stop_rfkill(struct iwl_trans *trans, iwl_trans_pcie_rf_kill(trans, hw_rfkill, false); } -static void iwl_trans_pcie_stop_device(struct iwl_trans *trans) +void iwl_trans_pcie_stop_device(struct iwl_trans *trans) { struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); bool was_in_rfkill; @@ -1505,9 +1504,17 @@ void iwl_pcie_d3_complete_suspend(struct iwl_trans *trans, iwl_pcie_synchronize_irqs(trans); - iwl_clear_bit(trans, CSR_GP_CNTRL, - CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ); - iwl_clear_bit(trans, CSR_GP_CNTRL, CSR_GP_CNTRL_REG_FLAG_INIT_DONE); + if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_BZ) { + iwl_clear_bit(trans, CSR_GP_CNTRL, + CSR_GP_CNTRL_REG_FLAG_BZ_MAC_ACCESS_REQ); + iwl_clear_bit(trans, CSR_GP_CNTRL, + CSR_GP_CNTRL_REG_FLAG_MAC_INIT); + } else { + iwl_clear_bit(trans, CSR_GP_CNTRL, + CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ); + iwl_clear_bit(trans, CSR_GP_CNTRL, + CSR_GP_CNTRL_REG_FLAG_INIT_DONE); + } if (reset) { /* @@ -1552,8 +1559,7 @@ static int iwl_pcie_d3_handshake(struct iwl_trans *trans, bool suspend) return 0; } -static int iwl_trans_pcie_d3_suspend(struct iwl_trans *trans, bool test, - bool reset) +int iwl_trans_pcie_d3_suspend(struct iwl_trans *trans, bool test, bool reset) { int ret; @@ -1571,9 +1577,9 @@ static int iwl_trans_pcie_d3_suspend(struct iwl_trans *trans, bool test, return 0; } -static int iwl_trans_pcie_d3_resume(struct iwl_trans *trans, - enum iwl_d3_status *status, - bool test, bool reset) +int iwl_trans_pcie_d3_resume(struct iwl_trans *trans, + enum iwl_d3_status *status, + bool test, bool reset) { struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); u32 val; @@ -1586,8 +1592,12 @@ static int iwl_trans_pcie_d3_resume(struct iwl_trans *trans, goto out; } - iwl_set_bit(trans, CSR_GP_CNTRL, - CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ); + if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_BZ) + iwl_set_bit(trans, CSR_GP_CNTRL, + CSR_GP_CNTRL_REG_FLAG_BZ_MAC_ACCESS_REQ); + else + iwl_set_bit(trans, CSR_GP_CNTRL, + CSR_GP_CNTRL_REG_FLAG_MAC_ACCESS_REQ); ret = iwl_finish_nic_init(trans); if (ret) @@ -1874,7 +1884,7 @@ static int _iwl_trans_pcie_start_hw(struct iwl_trans *trans) return 0; } -static int iwl_trans_pcie_start_hw(struct iwl_trans *trans) +int iwl_trans_pcie_start_hw(struct iwl_trans *trans) { struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); int ret; @@ -1886,7 +1896,7 @@ static int iwl_trans_pcie_start_hw(struct iwl_trans *trans) return ret; } -static void iwl_trans_pcie_op_mode_leave(struct iwl_trans *trans) +void iwl_trans_pcie_op_mode_leave(struct iwl_trans *trans) { struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); @@ -1906,17 +1916,17 @@ static void iwl_trans_pcie_op_mode_leave(struct iwl_trans *trans) iwl_pcie_synchronize_irqs(trans); } -static void iwl_trans_pcie_write8(struct iwl_trans *trans, u32 ofs, u8 val) +void iwl_trans_pcie_write8(struct iwl_trans *trans, u32 ofs, u8 val) { writeb(val, IWL_TRANS_GET_PCIE_TRANS(trans)->hw_base + ofs); } -static void iwl_trans_pcie_write32(struct iwl_trans *trans, u32 ofs, u32 val) +void iwl_trans_pcie_write32(struct iwl_trans *trans, u32 ofs, u32 val) { writel(val, IWL_TRANS_GET_PCIE_TRANS(trans)->hw_base + ofs); } -static u32 iwl_trans_pcie_read32(struct iwl_trans *trans, u32 ofs) +u32 iwl_trans_pcie_read32(struct iwl_trans *trans, u32 ofs) { return readl(IWL_TRANS_GET_PCIE_TRANS(trans)->hw_base + ofs); } @@ -1929,7 +1939,7 @@ static u32 iwl_trans_pcie_prph_msk(struct iwl_trans *trans) return 0x000FFFFF; } -static u32 iwl_trans_pcie_read_prph(struct iwl_trans *trans, u32 reg) +u32 iwl_trans_pcie_read_prph(struct iwl_trans *trans, u32 reg) { u32 mask = iwl_trans_pcie_prph_msk(trans); @@ -1938,8 +1948,7 @@ static u32 iwl_trans_pcie_read_prph(struct iwl_trans *trans, u32 reg) return iwl_trans_pcie_read32(trans, HBUS_TARG_PRPH_RDAT); } -static void iwl_trans_pcie_write_prph(struct iwl_trans *trans, u32 addr, - u32 val) +void iwl_trans_pcie_write_prph(struct iwl_trans *trans, u32 addr, u32 val) { u32 mask = iwl_trans_pcie_prph_msk(trans); @@ -1948,20 +1957,19 @@ static void iwl_trans_pcie_write_prph(struct iwl_trans *trans, u32 addr, iwl_trans_pcie_write32(trans, HBUS_TARG_PRPH_WDAT, val); } -static void iwl_trans_pcie_configure(struct iwl_trans *trans, - const struct iwl_trans_config *trans_cfg) +void iwl_trans_pcie_configure(struct iwl_trans *trans, + const struct iwl_trans_config *trans_cfg) { struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); /* free all first - we might be reconfigured for a different size */ iwl_pcie_free_rbs_pool(trans); - trans->txqs.cmd.q_id = trans_cfg->cmd_queue; - trans->txqs.cmd.fifo = trans_cfg->cmd_fifo; - trans->txqs.cmd.wdg_timeout = trans_cfg->cmd_q_wdg_timeout; - trans->txqs.page_offs = trans_cfg->cb_data_offs; - trans->txqs.dev_cmd_offs = trans_cfg->cb_data_offs + sizeof(void *); - trans->txqs.queue_alloc_cmd_ver = trans_cfg->queue_alloc_cmd_ver; + trans_pcie->txqs.cmd.q_id = trans_cfg->cmd_queue; + trans_pcie->txqs.cmd.fifo = trans_cfg->cmd_fifo; + trans_pcie->txqs.page_offs = trans_cfg->cb_data_offs; + trans_pcie->txqs.dev_cmd_offs = trans_cfg->cb_data_offs + sizeof(void *); + trans_pcie->txqs.queue_alloc_cmd_ver = trans_cfg->queue_alloc_cmd_ver; if (WARN_ON(trans_cfg->n_no_reclaim_cmds > MAX_NO_RECLAIM_CMDS)) trans_pcie->n_no_reclaim_cmds = 0; @@ -1980,19 +1988,12 @@ static void iwl_trans_pcie_configure(struct iwl_trans *trans, if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_AX210) trans_pcie->supported_dma_mask = DMA_BIT_MASK(11); - trans->txqs.bc_table_dword = trans_cfg->bc_table_dword; + trans_pcie->txqs.bc_table_dword = trans_cfg->bc_table_dword; trans_pcie->scd_set_active = trans_cfg->scd_set_active; trans->command_groups = trans_cfg->command_groups; trans->command_groups_size = trans_cfg->command_groups_size; - /* Initialize NAPI here - it should be before registering to mac80211 - * in the opmode but after the HW struct is allocated. - * As this function may be called again in some corner cases don't - * do anything if NAPI was already initialized. - */ - if (trans_pcie->napi_dev.reg_state != NETREG_DUMMY) - init_dummy_netdev(&trans_pcie->napi_dev); trans_pcie->fw_reset_handshake = trans_cfg->fw_reset_handshake; } @@ -2074,6 +2075,8 @@ void iwl_trans_pcie_free(struct iwl_trans *trans) iwl_pcie_free_ict(trans); } + free_netdev(trans_pcie->napi_dev); + iwl_pcie_free_invalid_tx_cmd(trans); iwl_pcie_free_fw_monitor(trans); @@ -2084,15 +2087,20 @@ void iwl_trans_pcie_free(struct iwl_trans *trans) trans->dev); mutex_destroy(&trans_pcie->mutex); - iwl_trans_free(trans); -} -static void iwl_trans_pcie_set_pmi(struct iwl_trans *trans, bool state) -{ - if (state) - set_bit(STATUS_TPOWER_PMI, &trans->status); - else - clear_bit(STATUS_TPOWER_PMI, &trans->status); + if (trans_pcie->txqs.tso_hdr_page) { + for_each_possible_cpu(i) { + struct iwl_tso_hdr_page *p = + per_cpu_ptr(trans_pcie->txqs.tso_hdr_page, i); + + if (p && p->page) + __free_page(p->page); + } + + free_percpu(trans_pcie->txqs.tso_hdr_page); + } + + iwl_trans_free(trans); } struct iwl_trans_pcie_removal { @@ -2258,7 +2266,7 @@ out: return true; } -static bool iwl_trans_pcie_grab_nic_access(struct iwl_trans *trans) +bool iwl_trans_pcie_grab_nic_access(struct iwl_trans *trans) { bool ret; @@ -2272,7 +2280,7 @@ static bool iwl_trans_pcie_grab_nic_access(struct iwl_trans *trans) return false; } -static void iwl_trans_pcie_release_nic_access(struct iwl_trans *trans) +void iwl_trans_pcie_release_nic_access(struct iwl_trans *trans) { struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); @@ -2302,8 +2310,8 @@ out: spin_unlock_bh(&trans_pcie->reg_lock); } -static int iwl_trans_pcie_read_mem(struct iwl_trans *trans, u32 addr, - void *buf, int dwords) +int iwl_trans_pcie_read_mem(struct iwl_trans *trans, u32 addr, + void *buf, int dwords) { #define IWL_MAX_HW_ERRS 5 unsigned int num_consec_hw_errors = 0; @@ -2352,8 +2360,8 @@ static int iwl_trans_pcie_read_mem(struct iwl_trans *trans, u32 addr, return 0; } -static int iwl_trans_pcie_write_mem(struct iwl_trans *trans, u32 addr, - const void *buf, int dwords) +int iwl_trans_pcie_write_mem(struct iwl_trans *trans, u32 addr, + const void *buf, int dwords) { int offs, ret = 0; const u32 *vals = buf; @@ -2370,8 +2378,8 @@ static int iwl_trans_pcie_write_mem(struct iwl_trans *trans, u32 addr, return ret; } -static int iwl_trans_pcie_read_config32(struct iwl_trans *trans, u32 ofs, - u32 *val) +int iwl_trans_pcie_read_config32(struct iwl_trans *trans, u32 ofs, + u32 *val) { return pci_read_config_dword(IWL_TRANS_GET_PCIE_TRANS(trans)->pci_dev, ofs, val); @@ -2379,8 +2387,8 @@ static int iwl_trans_pcie_read_config32(struct iwl_trans *trans, u32 ofs, #define IWL_FLUSH_WAIT_MS 2000 -static int iwl_trans_pcie_rxq_dma_data(struct iwl_trans *trans, int queue, - struct iwl_trans_rxq_dma_data *data) +int iwl_trans_pcie_rxq_dma_data(struct iwl_trans *trans, int queue, + struct iwl_trans_rxq_dma_data *data) { struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); @@ -2395,8 +2403,9 @@ static int iwl_trans_pcie_rxq_dma_data(struct iwl_trans *trans, int queue, return 0; } -static int iwl_trans_pcie_wait_txq_empty(struct iwl_trans *trans, int txq_idx) +int iwl_trans_pcie_wait_txq_empty(struct iwl_trans *trans, int txq_idx) { + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); struct iwl_txq *txq; unsigned long now = jiffies; bool overflow_tx; @@ -2406,11 +2415,11 @@ static int iwl_trans_pcie_wait_txq_empty(struct iwl_trans *trans, int txq_idx) if (test_bit(STATUS_TRANS_DEAD, &trans->status)) return -ENODEV; - if (!test_bit(txq_idx, trans->txqs.queue_used)) + if (!test_bit(txq_idx, trans_pcie->txqs.queue_used)) return -EINVAL; IWL_DEBUG_TX_QUEUES(trans, "Emptying queue %d...\n", txq_idx); - txq = trans->txqs.txq[txq_idx]; + txq = trans_pcie->txqs.txq[txq_idx]; spin_lock_bh(&txq->lock); overflow_tx = txq->overflow_tx || @@ -2456,8 +2465,9 @@ static int iwl_trans_pcie_wait_txq_empty(struct iwl_trans *trans, int txq_idx) return 0; } -static int iwl_trans_pcie_wait_txqs_empty(struct iwl_trans *trans, u32 txq_bm) +int iwl_trans_pcie_wait_txqs_empty(struct iwl_trans *trans, u32 txq_bm) { + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); int cnt; int ret = 0; @@ -2466,9 +2476,9 @@ static int iwl_trans_pcie_wait_txqs_empty(struct iwl_trans *trans, u32 txq_bm) cnt < trans->trans_cfg->base_params->num_of_queues; cnt++) { - if (cnt == trans->txqs.cmd.q_id) + if (cnt == trans_pcie->txqs.cmd.q_id) continue; - if (!test_bit(cnt, trans->txqs.queue_used)) + if (!test_bit(cnt, trans_pcie->txqs.queue_used)) continue; if (!(BIT(cnt) & txq_bm)) continue; @@ -2481,8 +2491,8 @@ static int iwl_trans_pcie_wait_txqs_empty(struct iwl_trans *trans, u32 txq_bm) return ret; } -static void iwl_trans_pcie_set_bits_mask(struct iwl_trans *trans, u32 reg, - u32 mask, u32 value) +void iwl_trans_pcie_set_bits_mask(struct iwl_trans *trans, u32 reg, + u32 mask, u32 value) { struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); @@ -2641,12 +2651,13 @@ static int iwl_dbgfs_tx_queue_seq_show(struct seq_file *seq, void *v) struct iwl_dbgfs_tx_queue_priv *priv = seq->private; struct iwl_dbgfs_tx_queue_state *state = v; struct iwl_trans *trans = priv->trans; - struct iwl_txq *txq = trans->txqs.txq[state->pos]; + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwl_txq *txq = trans_pcie->txqs.txq[state->pos]; seq_printf(seq, "hwq %.3u: used=%d stopped=%d ", (unsigned int)state->pos, - !!test_bit(state->pos, trans->txqs.queue_used), - !!test_bit(state->pos, trans->txqs.queue_stopped)); + !!test_bit(state->pos, trans_pcie->txqs.queue_used), + !!test_bit(state->pos, trans_pcie->txqs.queue_stopped)); if (txq) seq_printf(seq, "read=%u write=%u need_update=%d frozen=%d n_window=%d ampdu=%d", @@ -2656,7 +2667,7 @@ static int iwl_dbgfs_tx_queue_seq_show(struct seq_file *seq, void *v) else seq_puts(seq, "(unallocated)"); - if (state->pos == trans->txqs.cmd.q_id) + if (state->pos == trans_pcie->txqs.cmd.q_id) seq_puts(seq, " (HCMD)"); seq_puts(seq, "\n"); @@ -3060,7 +3071,7 @@ void iwl_trans_pcie_dbgfs_register(struct iwl_trans *trans) DEBUGFS_ADD_FILE(rf, dir, 0400); } -static void iwl_trans_pcie_debugfs_cleanup(struct iwl_trans *trans) +void iwl_trans_pcie_debugfs_cleanup(struct iwl_trans *trans) { struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); struct cont_rec *data = &trans_pcie->fw_mon_data; @@ -3073,10 +3084,11 @@ static void iwl_trans_pcie_debugfs_cleanup(struct iwl_trans *trans) static u32 iwl_trans_pcie_get_cmdlen(struct iwl_trans *trans, void *tfd) { + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); u32 cmdlen = 0; int i; - for (i = 0; i < trans->txqs.tfd.max_tbs; i++) + for (i = 0; i < trans_pcie->txqs.tfd.max_tbs; i++) cmdlen += iwl_txq_gen1_tfd_tb_get_len(trans, tfd, i); return cmdlen; @@ -3337,15 +3349,14 @@ static int iwl_trans_get_fw_monitor_len(struct iwl_trans *trans, u32 *len) return 0; } -static struct iwl_trans_dump_data * -iwl_trans_pcie_dump_data(struct iwl_trans *trans, - u32 dump_mask, +struct iwl_trans_dump_data * +iwl_trans_pcie_dump_data(struct iwl_trans *trans, u32 dump_mask, const struct iwl_dump_sanitize_ops *sanitize_ops, void *sanitize_ctx) { struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); struct iwl_fw_error_dump_data *data; - struct iwl_txq *cmdq = trans->txqs.txq[trans->txqs.cmd.q_id]; + struct iwl_txq *cmdq = trans_pcie->txqs.txq[trans_pcie->txqs.cmd.q_id]; struct iwl_fw_error_dump_txcmd *txcmd; struct iwl_trans_dump_data *dump_data; u32 len, num_rbs = 0, monitor_len = 0; @@ -3412,7 +3423,7 @@ iwl_trans_pcie_dump_data(struct iwl_trans *trans, data = (void *)dump_data->data; if (dump_mask & BIT(IWL_FW_ERROR_DUMP_TXCMD) && cmdq) { - u16 tfd_size = trans->txqs.tfd.size; + u16 tfd_size = trans_pcie->txqs.tfd.size; data->type = cpu_to_le32(IWL_FW_ERROR_DUMP_TXCMD); txcmd = (void *)data->data; @@ -3488,7 +3499,7 @@ iwl_trans_pcie_dump_data(struct iwl_trans *trans, return dump_data; } -static void iwl_trans_pci_interrupts(struct iwl_trans *trans, bool enable) +void iwl_trans_pci_interrupts(struct iwl_trans *trans, bool enable) { if (enable) iwl_enable_interrupts(trans); @@ -3496,7 +3507,7 @@ static void iwl_trans_pci_interrupts(struct iwl_trans *trans, bool enable) iwl_disable_interrupts(trans); } -static void iwl_trans_pcie_sync_nmi(struct iwl_trans *trans) +void iwl_trans_pcie_sync_nmi(struct iwl_trans *trans) { u32 inta_addr, sw_err_bit; struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); @@ -3515,95 +3526,16 @@ static void iwl_trans_pcie_sync_nmi(struct iwl_trans *trans) iwl_trans_sync_nmi_with_addr(trans, inta_addr, sw_err_bit); } -#define IWL_TRANS_COMMON_OPS \ - .op_mode_leave = iwl_trans_pcie_op_mode_leave, \ - .write8 = iwl_trans_pcie_write8, \ - .write32 = iwl_trans_pcie_write32, \ - .read32 = iwl_trans_pcie_read32, \ - .read_prph = iwl_trans_pcie_read_prph, \ - .write_prph = iwl_trans_pcie_write_prph, \ - .read_mem = iwl_trans_pcie_read_mem, \ - .write_mem = iwl_trans_pcie_write_mem, \ - .read_config32 = iwl_trans_pcie_read_config32, \ - .configure = iwl_trans_pcie_configure, \ - .set_pmi = iwl_trans_pcie_set_pmi, \ - .sw_reset = iwl_trans_pcie_sw_reset, \ - .grab_nic_access = iwl_trans_pcie_grab_nic_access, \ - .release_nic_access = iwl_trans_pcie_release_nic_access, \ - .set_bits_mask = iwl_trans_pcie_set_bits_mask, \ - .dump_data = iwl_trans_pcie_dump_data, \ - .d3_suspend = iwl_trans_pcie_d3_suspend, \ - .d3_resume = iwl_trans_pcie_d3_resume, \ - .interrupts = iwl_trans_pci_interrupts, \ - .sync_nmi = iwl_trans_pcie_sync_nmi, \ - .imr_dma_data = iwl_trans_pcie_copy_imr \ - -static const struct iwl_trans_ops trans_ops_pcie = { - IWL_TRANS_COMMON_OPS, - .start_hw = iwl_trans_pcie_start_hw, - .fw_alive = iwl_trans_pcie_fw_alive, - .start_fw = iwl_trans_pcie_start_fw, - .stop_device = iwl_trans_pcie_stop_device, - - .send_cmd = iwl_pcie_enqueue_hcmd, - - .tx = iwl_trans_pcie_tx, - .reclaim = iwl_txq_reclaim, - - .txq_disable = iwl_trans_pcie_txq_disable, - .txq_enable = iwl_trans_pcie_txq_enable, - - .txq_set_shared_mode = iwl_trans_pcie_txq_set_shared_mode, - - .wait_tx_queues_empty = iwl_trans_pcie_wait_txqs_empty, - - .freeze_txq_timer = iwl_trans_txq_freeze_timer, -#ifdef CONFIG_IWLWIFI_DEBUGFS - .debugfs_cleanup = iwl_trans_pcie_debugfs_cleanup, -#endif -}; - -static const struct iwl_trans_ops trans_ops_pcie_gen2 = { - IWL_TRANS_COMMON_OPS, - .start_hw = iwl_trans_pcie_start_hw, - .fw_alive = iwl_trans_pcie_gen2_fw_alive, - .start_fw = iwl_trans_pcie_gen2_start_fw, - .stop_device = iwl_trans_pcie_gen2_stop_device, - - .send_cmd = iwl_pcie_gen2_enqueue_hcmd, - - .tx = iwl_txq_gen2_tx, - .reclaim = iwl_txq_reclaim, - - .set_q_ptrs = iwl_txq_set_q_ptrs, - - .txq_alloc = iwl_txq_dyn_alloc, - .txq_free = iwl_txq_dyn_free, - .wait_txq_empty = iwl_trans_pcie_wait_txq_empty, - .rxq_dma_data = iwl_trans_pcie_rxq_dma_data, - .load_pnvm = iwl_trans_pcie_ctx_info_gen3_load_pnvm, - .set_pnvm = iwl_trans_pcie_ctx_info_gen3_set_pnvm, - .load_reduce_power = iwl_trans_pcie_ctx_info_gen3_load_reduce_power, - .set_reduce_power = iwl_trans_pcie_ctx_info_gen3_set_reduce_power, -#ifdef CONFIG_IWLWIFI_DEBUGFS - .debugfs_cleanup = iwl_trans_pcie_debugfs_cleanup, -#endif -}; - struct iwl_trans *iwl_trans_pcie_alloc(struct pci_dev *pdev, const struct pci_device_id *ent, const struct iwl_cfg_trans_params *cfg_trans) { - struct iwl_trans_pcie *trans_pcie; + struct iwl_trans_pcie *trans_pcie, **priv; struct iwl_trans *trans; int ret, addr_size; - const struct iwl_trans_ops *ops = &trans_ops_pcie_gen2; void __iomem * const *table; u32 bar0; - if (!cfg_trans->gen2) - ops = &trans_ops_pcie; - /* reassign our BAR 0 if invalid due to possible runtime PM races */ pci_read_config_dword(pdev, PCI_BASE_ADDRESS_0, &bar0); if (bar0 == PCI_BASE_ADDRESS_MEM_TYPE_64) { @@ -3616,13 +3548,73 @@ struct iwl_trans *iwl_trans_pcie_alloc(struct pci_dev *pdev, if (ret) return ERR_PTR(ret); - trans = iwl_trans_alloc(sizeof(struct iwl_trans_pcie), &pdev->dev, ops, + trans = iwl_trans_alloc(sizeof(struct iwl_trans_pcie), &pdev->dev, cfg_trans); if (!trans) return ERR_PTR(-ENOMEM); trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + if (trans->trans_cfg->gen2) { + trans_pcie->txqs.tfd.addr_size = 64; + trans_pcie->txqs.tfd.max_tbs = IWL_TFH_NUM_TBS; + trans_pcie->txqs.tfd.size = sizeof(struct iwl_tfh_tfd); + } else { + trans_pcie->txqs.tfd.addr_size = 36; + trans_pcie->txqs.tfd.max_tbs = IWL_NUM_OF_TBS; + trans_pcie->txqs.tfd.size = sizeof(struct iwl_tfd); + } + trans->max_skb_frags = IWL_TRANS_PCIE_MAX_FRAGS(trans_pcie); + + /* Set a short watchdog for the command queue */ + trans_pcie->txqs.cmd.wdg_timeout = IWL_DEF_WD_TIMEOUT; + + trans_pcie->txqs.tso_hdr_page = alloc_percpu(struct iwl_tso_hdr_page); + if (!trans_pcie->txqs.tso_hdr_page) { + ret = -ENOMEM; + goto out_free_trans; + } + + if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_BZ) + trans_pcie->txqs.bc_tbl_size = + sizeof(struct iwl_gen3_bc_tbl_entry) * TFD_QUEUE_BC_SIZE_GEN3_BZ; + else if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_AX210) + trans_pcie->txqs.bc_tbl_size = + sizeof(struct iwl_gen3_bc_tbl_entry) * TFD_QUEUE_BC_SIZE_GEN3_AX210; + else + trans_pcie->txqs.bc_tbl_size = sizeof(struct iwlagn_scd_bc_tbl); + /* + * For gen2 devices, we use a single allocation for each byte-count + * table, but they're pretty small (1k) so use a DMA pool that we + * allocate here. + */ + if (trans->trans_cfg->gen2) { + trans_pcie->txqs.bc_pool = + dmam_pool_create("iwlwifi:bc", trans->dev, + trans_pcie->txqs.bc_tbl_size, + 256, 0); + if (!trans_pcie->txqs.bc_pool) { + ret = -ENOMEM; + goto out_free_tso; + } + } + + /* Some things must not change even if the config does */ + WARN_ON(trans_pcie->txqs.tfd.addr_size != + (trans->trans_cfg->gen2 ? 64 : 36)); + + /* Initialize NAPI here - it should be before registering to mac80211 + * in the opmode but after the HW struct is allocated. + */ + trans_pcie->napi_dev = alloc_netdev_dummy(sizeof(struct iwl_trans_pcie *)); + if (!trans_pcie->napi_dev) { + ret = -ENOMEM; + goto out_free_tso; + } + /* The private struct in netdev is a pointer to struct iwl_trans_pcie */ + priv = netdev_priv(trans_pcie->napi_dev); + *priv = trans_pcie; + trans_pcie->trans = trans; trans_pcie->opmode_down = true; spin_lock_init(&trans_pcie->irq_lock); @@ -3637,7 +3629,7 @@ struct iwl_trans *iwl_trans_pcie_alloc(struct pci_dev *pdev, WQ_HIGHPRI | WQ_UNBOUND, 0); if (!trans_pcie->rba.alloc_wq) { ret = -ENOMEM; - goto out_free_trans; + goto out_free_ndev; } INIT_WORK(&trans_pcie->rba.rx_alloc, iwl_pcie_rx_allocator_work); @@ -3656,7 +3648,7 @@ struct iwl_trans *iwl_trans_pcie_alloc(struct pci_dev *pdev, pci_set_master(pdev); - addr_size = trans->txqs.tfd.addr_size; + addr_size = trans_pcie->txqs.tfd.addr_size; ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(addr_size)); if (ret) { ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)); @@ -3757,6 +3749,10 @@ out_free_ict: iwl_pcie_free_ict(trans); out_no_pci: destroy_workqueue(trans_pcie->rba.alloc_wq); +out_free_ndev: + free_netdev(trans_pcie->napi_dev); +out_free_tso: + free_percpu(trans_pcie->txqs.tso_hdr_page); out_free_trans: iwl_trans_free(trans); return ERR_PTR(ret); diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/tx-gen2.c b/drivers/net/wireless/intel/iwlwifi/pcie/tx-gen2.c index aabbef114bc2..b1846abb99b7 100644 --- a/drivers/net/wireless/intel/iwlwifi/pcie/tx-gen2.c +++ b/drivers/net/wireless/intel/iwlwifi/pcie/tx-gen2.c @@ -1,7 +1,7 @@ // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause /* * Copyright (C) 2017 Intel Deutschland GmbH - * Copyright (C) 2018-2020, 2023 Intel Corporation + * Copyright (C) 2018-2020, 2023-2024 Intel Corporation */ #include <net/tso.h> #include <linux/tcp.h> @@ -11,7 +11,1183 @@ #include "iwl-io.h" #include "internal.h" #include "fw/api/tx.h" -#include "queue/tx.h" +#include "fw/api/commands.h" +#include "fw/api/datapath.h" +#include "iwl-scd.h" + +static struct page *get_workaround_page(struct iwl_trans *trans, + struct sk_buff *skb) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwl_tso_page_info *info; + struct page **page_ptr; + struct page *ret; + dma_addr_t phys; + + page_ptr = (void *)((u8 *)skb->cb + trans_pcie->txqs.page_offs); + + ret = alloc_page(GFP_ATOMIC); + if (!ret) + return NULL; + + info = IWL_TSO_PAGE_INFO(page_address(ret)); + + /* Create a DMA mapping for the page */ + phys = dma_map_page_attrs(trans->dev, ret, 0, PAGE_SIZE, + DMA_TO_DEVICE, DMA_ATTR_SKIP_CPU_SYNC); + if (unlikely(dma_mapping_error(trans->dev, phys))) { + __free_page(ret); + return NULL; + } + + /* Store physical address and set use count */ + info->dma_addr = phys; + refcount_set(&info->use_count, 1); + + /* set the chaining pointer to the previous page if there */ + info->next = *page_ptr; + *page_ptr = ret; + + return ret; +} + +/* + * Add a TB and if needed apply the FH HW bug workaround; + * meta != NULL indicates that it's a page mapping and we + * need to dma_unmap_page() and set the meta->tbs bit in + * this case. + */ +static int iwl_txq_gen2_set_tb_with_wa(struct iwl_trans *trans, + struct sk_buff *skb, + struct iwl_tfh_tfd *tfd, + dma_addr_t phys, void *virt, + u16 len, struct iwl_cmd_meta *meta, + bool unmap) +{ + dma_addr_t oldphys = phys; + struct page *page; + int ret; + + if (unlikely(dma_mapping_error(trans->dev, phys))) + return -ENOMEM; + + if (likely(!iwl_txq_crosses_4g_boundary(phys, len))) { + ret = iwl_txq_gen2_set_tb(trans, tfd, phys, len); + + if (ret < 0) + goto unmap; + + if (meta) + meta->tbs |= BIT(ret); + + ret = 0; + goto trace; + } + + /* + * Work around a hardware bug. If (as expressed in the + * condition above) the TB ends on a 32-bit boundary, + * then the next TB may be accessed with the wrong + * address. + * To work around it, copy the data elsewhere and make + * a new mapping for it so the device will not fail. + */ + + if (WARN_ON(len > IWL_TSO_PAGE_DATA_SIZE)) { + ret = -ENOBUFS; + goto unmap; + } + + page = get_workaround_page(trans, skb); + if (!page) { + ret = -ENOMEM; + goto unmap; + } + + memcpy(page_address(page), virt, len); + + /* + * This is a bit odd, but performance does not matter here, what + * matters are the expectations of the calling code and TB cleanup + * function. + * + * As such, if unmap is set, then create another mapping for the TB + * entry as it will be unmapped later. On the other hand, if it is not + * set, then the TB entry will not be unmapped and instead we simply + * reference and sync the mapping that get_workaround_page() created. + */ + if (unmap) { + phys = dma_map_single(trans->dev, page_address(page), len, + DMA_TO_DEVICE); + if (unlikely(dma_mapping_error(trans->dev, phys))) + return -ENOMEM; + } else { + phys = iwl_pcie_get_tso_page_phys(page_address(page)); + dma_sync_single_for_device(trans->dev, phys, len, + DMA_TO_DEVICE); + } + + ret = iwl_txq_gen2_set_tb(trans, tfd, phys, len); + if (ret < 0) { + /* unmap the new allocation as single */ + oldphys = phys; + meta = NULL; + goto unmap; + } + + IWL_DEBUG_TX(trans, + "TB bug workaround: copied %d bytes from 0x%llx to 0x%llx\n", + len, (unsigned long long)oldphys, + (unsigned long long)phys); + + ret = 0; +unmap: + if (!unmap) + goto trace; + + if (meta) + dma_unmap_page(trans->dev, oldphys, len, DMA_TO_DEVICE); + else + dma_unmap_single(trans->dev, oldphys, len, DMA_TO_DEVICE); +trace: + trace_iwlwifi_dev_tx_tb(trans->dev, skb, virt, phys, len); + + return ret; +} + +static int iwl_txq_gen2_build_amsdu(struct iwl_trans *trans, + struct sk_buff *skb, + struct iwl_tfh_tfd *tfd, + struct iwl_cmd_meta *out_meta, + int start_len, + u8 hdr_len, + struct iwl_device_tx_cmd *dev_cmd) +{ +#ifdef CONFIG_INET + struct iwl_tx_cmd_gen2 *tx_cmd = (void *)dev_cmd->payload; + struct ieee80211_hdr *hdr = (void *)skb->data; + unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room; + unsigned int mss = skb_shinfo(skb)->gso_size; + unsigned int data_offset = 0; + dma_addr_t start_hdr_phys; + u16 length, amsdu_pad; + u8 *start_hdr; + struct sg_table *sgt; + struct tso_t tso; + + trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd), + &dev_cmd->hdr, start_len, 0); + + ip_hdrlen = skb_network_header_len(skb); + snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb); + total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len; + amsdu_pad = 0; + + /* total amount of header we may need for this A-MSDU */ + hdr_room = DIV_ROUND_UP(total_len, mss) * + (3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)); + + /* Our device supports 9 segments at most, it will fit in 1 page */ + sgt = iwl_pcie_prep_tso(trans, skb, out_meta, &start_hdr, hdr_room); + if (!sgt) + return -ENOMEM; + + start_hdr_phys = iwl_pcie_get_tso_page_phys(start_hdr); + + /* + * Pull the ieee80211 header to be able to use TSO core, + * we will restore it for the tx_status flow. + */ + skb_pull(skb, hdr_len); + + /* + * Remove the length of all the headers that we don't actually + * have in the MPDU by themselves, but that we duplicate into + * all the different MSDUs inside the A-MSDU. + */ + le16_add_cpu(&tx_cmd->len, -snap_ip_tcp_hdrlen); + + tso_start(skb, &tso); + + while (total_len) { + /* this is the data left for this subframe */ + unsigned int data_left = min_t(unsigned int, mss, total_len); + unsigned int tb_len; + dma_addr_t tb_phys; + u8 *pos_hdr = start_hdr; + + total_len -= data_left; + + memset(pos_hdr, 0, amsdu_pad); + pos_hdr += amsdu_pad; + amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen + + data_left)) & 0x3; + ether_addr_copy(pos_hdr, ieee80211_get_DA(hdr)); + pos_hdr += ETH_ALEN; + ether_addr_copy(pos_hdr, ieee80211_get_SA(hdr)); + pos_hdr += ETH_ALEN; + + length = snap_ip_tcp_hdrlen + data_left; + *((__be16 *)pos_hdr) = cpu_to_be16(length); + pos_hdr += sizeof(length); + + /* + * This will copy the SNAP as well which will be considered + * as MAC header. + */ + tso_build_hdr(skb, pos_hdr, &tso, data_left, !total_len); + + pos_hdr += snap_ip_tcp_hdrlen; + + tb_len = pos_hdr - start_hdr; + tb_phys = iwl_pcie_get_tso_page_phys(start_hdr); + + /* + * No need for _with_wa, this is from the TSO page and + * we leave some space at the end of it so can't hit + * the buggy scenario. + */ + iwl_txq_gen2_set_tb(trans, tfd, tb_phys, tb_len); + trace_iwlwifi_dev_tx_tb(trans->dev, skb, start_hdr, + tb_phys, tb_len); + /* add this subframe's headers' length to the tx_cmd */ + le16_add_cpu(&tx_cmd->len, tb_len); + + /* prepare the start_hdr for the next subframe */ + start_hdr = pos_hdr; + + /* put the payload */ + while (data_left) { + int ret; + + tb_len = min_t(unsigned int, tso.size, data_left); + tb_phys = iwl_pcie_get_sgt_tb_phys(sgt, data_offset, + tb_len); + /* Not a real mapping error, use direct comparison */ + if (unlikely(tb_phys == DMA_MAPPING_ERROR)) + goto out_err; + + ret = iwl_txq_gen2_set_tb_with_wa(trans, skb, tfd, + tb_phys, tso.data, + tb_len, NULL, false); + if (ret) + goto out_err; + + data_left -= tb_len; + data_offset += tb_len; + tso_build_data(skb, &tso, tb_len); + } + } + + dma_sync_single_for_device(trans->dev, start_hdr_phys, hdr_room, + DMA_TO_DEVICE); + + /* re -add the WiFi header */ + skb_push(skb, hdr_len); + + return 0; + +out_err: +#endif + return -EINVAL; +} + +static struct +iwl_tfh_tfd *iwl_txq_gen2_build_tx_amsdu(struct iwl_trans *trans, + struct iwl_txq *txq, + struct iwl_device_tx_cmd *dev_cmd, + struct sk_buff *skb, + struct iwl_cmd_meta *out_meta, + int hdr_len, + int tx_cmd_len) +{ + int idx = iwl_txq_get_cmd_index(txq, txq->write_ptr); + struct iwl_tfh_tfd *tfd = iwl_txq_get_tfd(trans, txq, idx); + dma_addr_t tb_phys; + int len; + void *tb1_addr; + + tb_phys = iwl_txq_get_first_tb_dma(txq, idx); + + /* + * No need for _with_wa, the first TB allocation is aligned up + * to a 64-byte boundary and thus can't be at the end or cross + * a page boundary (much less a 2^32 boundary). + */ + iwl_txq_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE); + + /* + * The second TB (tb1) points to the remainder of the TX command + * and the 802.11 header - dword aligned size + * (This calculation modifies the TX command, so do it before the + * setup of the first TB) + */ + len = tx_cmd_len + sizeof(struct iwl_cmd_header) + hdr_len - + IWL_FIRST_TB_SIZE; + + /* do not align A-MSDU to dword as the subframe header aligns it */ + + /* map the data for TB1 */ + tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE; + tb_phys = dma_map_single(trans->dev, tb1_addr, len, DMA_TO_DEVICE); + if (unlikely(dma_mapping_error(trans->dev, tb_phys))) + goto out_err; + /* + * No need for _with_wa(), we ensure (via alignment) that the data + * here can never cross or end at a page boundary. + */ + iwl_txq_gen2_set_tb(trans, tfd, tb_phys, len); + + if (iwl_txq_gen2_build_amsdu(trans, skb, tfd, out_meta, + len + IWL_FIRST_TB_SIZE, hdr_len, dev_cmd)) + goto out_err; + + /* building the A-MSDU might have changed this data, memcpy it now */ + memcpy(&txq->first_tb_bufs[idx], dev_cmd, IWL_FIRST_TB_SIZE); + return tfd; + +out_err: + iwl_txq_gen2_tfd_unmap(trans, out_meta, tfd); + return NULL; +} + +static int iwl_txq_gen2_tx_add_frags(struct iwl_trans *trans, + struct sk_buff *skb, + struct iwl_tfh_tfd *tfd, + struct iwl_cmd_meta *out_meta) +{ + int i; + + for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { + const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; + dma_addr_t tb_phys; + unsigned int fragsz = skb_frag_size(frag); + int ret; + + if (!fragsz) + continue; + + tb_phys = skb_frag_dma_map(trans->dev, frag, 0, + fragsz, DMA_TO_DEVICE); + ret = iwl_txq_gen2_set_tb_with_wa(trans, skb, tfd, tb_phys, + skb_frag_address(frag), + fragsz, out_meta, true); + if (ret) + return ret; + } + + return 0; +} + +static struct +iwl_tfh_tfd *iwl_txq_gen2_build_tx(struct iwl_trans *trans, + struct iwl_txq *txq, + struct iwl_device_tx_cmd *dev_cmd, + struct sk_buff *skb, + struct iwl_cmd_meta *out_meta, + int hdr_len, + int tx_cmd_len, + bool pad) +{ + int idx = iwl_txq_get_cmd_index(txq, txq->write_ptr); + struct iwl_tfh_tfd *tfd = iwl_txq_get_tfd(trans, txq, idx); + dma_addr_t tb_phys; + int len, tb1_len, tb2_len; + void *tb1_addr; + struct sk_buff *frag; + + tb_phys = iwl_txq_get_first_tb_dma(txq, idx); + + /* The first TB points to bi-directional DMA data */ + memcpy(&txq->first_tb_bufs[idx], dev_cmd, IWL_FIRST_TB_SIZE); + + /* + * No need for _with_wa, the first TB allocation is aligned up + * to a 64-byte boundary and thus can't be at the end or cross + * a page boundary (much less a 2^32 boundary). + */ + iwl_txq_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE); + + /* + * The second TB (tb1) points to the remainder of the TX command + * and the 802.11 header - dword aligned size + * (This calculation modifies the TX command, so do it before the + * setup of the first TB) + */ + len = tx_cmd_len + sizeof(struct iwl_cmd_header) + hdr_len - + IWL_FIRST_TB_SIZE; + + if (pad) + tb1_len = ALIGN(len, 4); + else + tb1_len = len; + + /* map the data for TB1 */ + tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE; + tb_phys = dma_map_single(trans->dev, tb1_addr, tb1_len, DMA_TO_DEVICE); + if (unlikely(dma_mapping_error(trans->dev, tb_phys))) + goto out_err; + /* + * No need for _with_wa(), we ensure (via alignment) that the data + * here can never cross or end at a page boundary. + */ + iwl_txq_gen2_set_tb(trans, tfd, tb_phys, tb1_len); + trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd), &dev_cmd->hdr, + IWL_FIRST_TB_SIZE + tb1_len, hdr_len); + + /* set up TFD's third entry to point to remainder of skb's head */ + tb2_len = skb_headlen(skb) - hdr_len; + + if (tb2_len > 0) { + int ret; + + tb_phys = dma_map_single(trans->dev, skb->data + hdr_len, + tb2_len, DMA_TO_DEVICE); + ret = iwl_txq_gen2_set_tb_with_wa(trans, skb, tfd, tb_phys, + skb->data + hdr_len, tb2_len, + NULL, true); + if (ret) + goto out_err; + } + + if (iwl_txq_gen2_tx_add_frags(trans, skb, tfd, out_meta)) + goto out_err; + + skb_walk_frags(skb, frag) { + int ret; + + tb_phys = dma_map_single(trans->dev, frag->data, + skb_headlen(frag), DMA_TO_DEVICE); + ret = iwl_txq_gen2_set_tb_with_wa(trans, skb, tfd, tb_phys, + frag->data, + skb_headlen(frag), NULL, + true); + if (ret) + goto out_err; + if (iwl_txq_gen2_tx_add_frags(trans, frag, tfd, out_meta)) + goto out_err; + } + + return tfd; + +out_err: + iwl_txq_gen2_tfd_unmap(trans, out_meta, tfd); + return NULL; +} + +static +struct iwl_tfh_tfd *iwl_txq_gen2_build_tfd(struct iwl_trans *trans, + struct iwl_txq *txq, + struct iwl_device_tx_cmd *dev_cmd, + struct sk_buff *skb, + struct iwl_cmd_meta *out_meta) +{ + struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data; + int idx = iwl_txq_get_cmd_index(txq, txq->write_ptr); + struct iwl_tfh_tfd *tfd = iwl_txq_get_tfd(trans, txq, idx); + int len, hdr_len; + bool amsdu; + + /* There must be data left over for TB1 or this code must be changed */ + BUILD_BUG_ON(sizeof(struct iwl_tx_cmd_gen2) < IWL_FIRST_TB_SIZE); + BUILD_BUG_ON(sizeof(struct iwl_cmd_header) + + offsetofend(struct iwl_tx_cmd_gen2, dram_info) > + IWL_FIRST_TB_SIZE); + BUILD_BUG_ON(sizeof(struct iwl_tx_cmd_gen3) < IWL_FIRST_TB_SIZE); + BUILD_BUG_ON(sizeof(struct iwl_cmd_header) + + offsetofend(struct iwl_tx_cmd_gen3, dram_info) > + IWL_FIRST_TB_SIZE); + + memset(tfd, 0, sizeof(*tfd)); + + if (trans->trans_cfg->device_family < IWL_DEVICE_FAMILY_AX210) + len = sizeof(struct iwl_tx_cmd_gen2); + else + len = sizeof(struct iwl_tx_cmd_gen3); + + amsdu = ieee80211_is_data_qos(hdr->frame_control) && + (*ieee80211_get_qos_ctl(hdr) & + IEEE80211_QOS_CTL_A_MSDU_PRESENT); + + hdr_len = ieee80211_hdrlen(hdr->frame_control); + + /* + * Only build A-MSDUs here if doing so by GSO, otherwise it may be + * an A-MSDU for other reasons, e.g. NAN or an A-MSDU having been + * built in the higher layers already. + */ + if (amsdu && skb_shinfo(skb)->gso_size) + return iwl_txq_gen2_build_tx_amsdu(trans, txq, dev_cmd, skb, + out_meta, hdr_len, len); + return iwl_txq_gen2_build_tx(trans, txq, dev_cmd, skb, out_meta, + hdr_len, len, !amsdu); +} + +int iwl_txq_space(struct iwl_trans *trans, const struct iwl_txq *q) +{ + unsigned int max; + unsigned int used; + + /* + * To avoid ambiguity between empty and completely full queues, there + * should always be less than max_tfd_queue_size elements in the queue. + * If q->n_window is smaller than max_tfd_queue_size, there is no need + * to reserve any queue entries for this purpose. + */ + if (q->n_window < trans->trans_cfg->base_params->max_tfd_queue_size) + max = q->n_window; + else + max = trans->trans_cfg->base_params->max_tfd_queue_size - 1; + + /* + * max_tfd_queue_size is a power of 2, so the following is equivalent to + * modulo by max_tfd_queue_size and is well defined. + */ + used = (q->write_ptr - q->read_ptr) & + (trans->trans_cfg->base_params->max_tfd_queue_size - 1); + + if (WARN_ON(used > max)) + return 0; + + return max - used; +} + +/* + * iwl_pcie_gen2_update_byte_tbl - Set up entry in Tx byte-count array + */ +static void iwl_pcie_gen2_update_byte_tbl(struct iwl_trans *trans, + struct iwl_txq *txq, u16 byte_cnt, + int num_tbs) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + int idx = iwl_txq_get_cmd_index(txq, txq->write_ptr); + u8 filled_tfd_size, num_fetch_chunks; + u16 len = byte_cnt; + __le16 bc_ent; + + if (WARN(idx >= txq->n_window, "%d >= %d\n", idx, txq->n_window)) + return; + + filled_tfd_size = offsetof(struct iwl_tfh_tfd, tbs) + + num_tbs * sizeof(struct iwl_tfh_tb); + /* + * filled_tfd_size contains the number of filled bytes in the TFD. + * Dividing it by 64 will give the number of chunks to fetch + * to SRAM- 0 for one chunk, 1 for 2 and so on. + * If, for example, TFD contains only 3 TBs then 32 bytes + * of the TFD are used, and only one chunk of 64 bytes should + * be fetched + */ + num_fetch_chunks = DIV_ROUND_UP(filled_tfd_size, 64) - 1; + + if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_AX210) { + struct iwl_gen3_bc_tbl_entry *scd_bc_tbl_gen3 = txq->bc_tbl.addr; + + /* Starting from AX210, the HW expects bytes */ + WARN_ON(trans_pcie->txqs.bc_table_dword); + WARN_ON(len > 0x3FFF); + bc_ent = cpu_to_le16(len | (num_fetch_chunks << 14)); + scd_bc_tbl_gen3[idx].tfd_offset = bc_ent; + } else { + struct iwlagn_scd_bc_tbl *scd_bc_tbl = txq->bc_tbl.addr; + + /* Before AX210, the HW expects DW */ + WARN_ON(!trans_pcie->txqs.bc_table_dword); + len = DIV_ROUND_UP(len, 4); + WARN_ON(len > 0xFFF); + bc_ent = cpu_to_le16(len | (num_fetch_chunks << 12)); + scd_bc_tbl->tfd_offset[idx] = bc_ent; + } +} + +static u8 iwl_txq_gen2_get_num_tbs(struct iwl_tfh_tfd *tfd) +{ + return le16_to_cpu(tfd->num_tbs) & 0x1f; +} + +int iwl_txq_gen2_set_tb(struct iwl_trans *trans, struct iwl_tfh_tfd *tfd, + dma_addr_t addr, u16 len) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + int idx = iwl_txq_gen2_get_num_tbs(tfd); + struct iwl_tfh_tb *tb; + + /* Only WARN here so we know about the issue, but we mess up our + * unmap path because not every place currently checks for errors + * returned from this function - it can only return an error if + * there's no more space, and so when we know there is enough we + * don't always check ... + */ + WARN(iwl_txq_crosses_4g_boundary(addr, len), + "possible DMA problem with iova:0x%llx, len:%d\n", + (unsigned long long)addr, len); + + if (WARN_ON(idx >= IWL_TFH_NUM_TBS)) + return -EINVAL; + tb = &tfd->tbs[idx]; + + /* Each TFD can point to a maximum max_tbs Tx buffers */ + if (le16_to_cpu(tfd->num_tbs) >= trans_pcie->txqs.tfd.max_tbs) { + IWL_ERR(trans, "Error can not send more than %d chunks\n", + trans_pcie->txqs.tfd.max_tbs); + return -EINVAL; + } + + put_unaligned_le64(addr, &tb->addr); + tb->tb_len = cpu_to_le16(len); + + tfd->num_tbs = cpu_to_le16(idx + 1); + + return idx; +} + +void iwl_txq_gen2_tfd_unmap(struct iwl_trans *trans, + struct iwl_cmd_meta *meta, + struct iwl_tfh_tfd *tfd) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + int i, num_tbs; + + /* Sanity check on number of chunks */ + num_tbs = iwl_txq_gen2_get_num_tbs(tfd); + + if (num_tbs > trans_pcie->txqs.tfd.max_tbs) { + IWL_ERR(trans, "Too many chunks: %i\n", num_tbs); + return; + } + + /* TB1 is mapped directly, the rest is the TSO page and SG list. */ + if (meta->sg_offset) + num_tbs = 2; + + /* first TB is never freed - it's the bidirectional DMA data */ + for (i = 1; i < num_tbs; i++) { + if (meta->tbs & BIT(i)) + dma_unmap_page(trans->dev, + le64_to_cpu(tfd->tbs[i].addr), + le16_to_cpu(tfd->tbs[i].tb_len), + DMA_TO_DEVICE); + else + dma_unmap_single(trans->dev, + le64_to_cpu(tfd->tbs[i].addr), + le16_to_cpu(tfd->tbs[i].tb_len), + DMA_TO_DEVICE); + } + + iwl_txq_set_tfd_invalid_gen2(trans, tfd); +} + +static void iwl_txq_gen2_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq) +{ + /* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and + * idx is bounded by n_window + */ + int idx = iwl_txq_get_cmd_index(txq, txq->read_ptr); + struct sk_buff *skb; + + lockdep_assert_held(&txq->lock); + + if (!txq->entries) + return; + + iwl_txq_gen2_tfd_unmap(trans, &txq->entries[idx].meta, + iwl_txq_get_tfd(trans, txq, idx)); + + skb = txq->entries[idx].skb; + + /* Can be called from irqs-disabled context + * If skb is not NULL, it means that the whole queue is being + * freed and that the queue is not empty - free the skb + */ + if (skb) { + iwl_op_mode_free_skb(trans->op_mode, skb); + txq->entries[idx].skb = NULL; + } +} + +/* + * iwl_txq_inc_wr_ptr - Send new write index to hardware + */ +static void iwl_txq_inc_wr_ptr(struct iwl_trans *trans, struct iwl_txq *txq) +{ + lockdep_assert_held(&txq->lock); + + IWL_DEBUG_TX(trans, "Q:%d WR: 0x%x\n", txq->id, txq->write_ptr); + + /* + * if not in power-save mode, uCode will never sleep when we're + * trying to tx (during RFKILL, we're not trying to tx). + */ + iwl_write32(trans, HBUS_TARG_WRPTR, txq->write_ptr | (txq->id << 16)); +} + +int iwl_txq_gen2_tx(struct iwl_trans *trans, struct sk_buff *skb, + struct iwl_device_tx_cmd *dev_cmd, int txq_id) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwl_cmd_meta *out_meta; + struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; + u16 cmd_len; + int idx; + void *tfd; + + if (WARN_ONCE(txq_id >= IWL_MAX_TVQM_QUEUES, + "queue %d out of range", txq_id)) + return -EINVAL; + + if (WARN_ONCE(!test_bit(txq_id, trans_pcie->txqs.queue_used), + "TX on unused queue %d\n", txq_id)) + return -EINVAL; + + if (skb_is_nonlinear(skb) && + skb_shinfo(skb)->nr_frags > IWL_TRANS_PCIE_MAX_FRAGS(trans_pcie) && + __skb_linearize(skb)) + return -ENOMEM; + + spin_lock(&txq->lock); + + if (iwl_txq_space(trans, txq) < txq->high_mark) { + iwl_txq_stop(trans, txq); + + /* don't put the packet on the ring, if there is no room */ + if (unlikely(iwl_txq_space(trans, txq) < 3)) { + struct iwl_device_tx_cmd **dev_cmd_ptr; + + dev_cmd_ptr = (void *)((u8 *)skb->cb + + trans_pcie->txqs.dev_cmd_offs); + + *dev_cmd_ptr = dev_cmd; + __skb_queue_tail(&txq->overflow_q, skb); + spin_unlock(&txq->lock); + return 0; + } + } + + idx = iwl_txq_get_cmd_index(txq, txq->write_ptr); + + /* Set up driver data for this TFD */ + txq->entries[idx].skb = skb; + txq->entries[idx].cmd = dev_cmd; + + dev_cmd->hdr.sequence = + cpu_to_le16((u16)(QUEUE_TO_SEQ(txq_id) | + INDEX_TO_SEQ(idx))); + + /* Set up first empty entry in queue's array of Tx/cmd buffers */ + out_meta = &txq->entries[idx].meta; + memset(out_meta, 0, sizeof(*out_meta)); + + tfd = iwl_txq_gen2_build_tfd(trans, txq, dev_cmd, skb, out_meta); + if (!tfd) { + spin_unlock(&txq->lock); + return -1; + } + + if (trans->trans_cfg->device_family >= IWL_DEVICE_FAMILY_AX210) { + struct iwl_tx_cmd_gen3 *tx_cmd_gen3 = + (void *)dev_cmd->payload; + + cmd_len = le16_to_cpu(tx_cmd_gen3->len); + } else { + struct iwl_tx_cmd_gen2 *tx_cmd_gen2 = + (void *)dev_cmd->payload; + + cmd_len = le16_to_cpu(tx_cmd_gen2->len); + } + + /* Set up entry for this TFD in Tx byte-count array */ + iwl_pcie_gen2_update_byte_tbl(trans, txq, cmd_len, + iwl_txq_gen2_get_num_tbs(tfd)); + + /* start timer if queue currently empty */ + if (txq->read_ptr == txq->write_ptr && txq->wd_timeout) + mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout); + + /* Tell device the write index *just past* this latest filled TFD */ + txq->write_ptr = iwl_txq_inc_wrap(trans, txq->write_ptr); + iwl_txq_inc_wr_ptr(trans, txq); + /* + * At this point the frame is "transmitted" successfully + * and we will get a TX status notification eventually. + */ + spin_unlock(&txq->lock); + return 0; +} + +/*************** HOST COMMAND QUEUE FUNCTIONS *****/ + +/* + * iwl_txq_gen2_unmap - Unmap any remaining DMA mappings and free skb's + */ +static void iwl_txq_gen2_unmap(struct iwl_trans *trans, int txq_id) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; + + spin_lock_bh(&txq->reclaim_lock); + spin_lock(&txq->lock); + while (txq->write_ptr != txq->read_ptr) { + IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n", + txq_id, txq->read_ptr); + + if (txq_id != trans_pcie->txqs.cmd.q_id) { + int idx = iwl_txq_get_cmd_index(txq, txq->read_ptr); + struct iwl_cmd_meta *cmd_meta = &txq->entries[idx].meta; + struct sk_buff *skb = txq->entries[idx].skb; + + if (!WARN_ON_ONCE(!skb)) + iwl_pcie_free_tso_pages(trans, skb, cmd_meta); + } + iwl_txq_gen2_free_tfd(trans, txq); + txq->read_ptr = iwl_txq_inc_wrap(trans, txq->read_ptr); + } + + while (!skb_queue_empty(&txq->overflow_q)) { + struct sk_buff *skb = __skb_dequeue(&txq->overflow_q); + + iwl_op_mode_free_skb(trans->op_mode, skb); + } + + spin_unlock(&txq->lock); + spin_unlock_bh(&txq->reclaim_lock); + + /* just in case - this queue may have been stopped */ + iwl_trans_pcie_wake_queue(trans, txq); +} + +static void iwl_txq_gen2_free_memory(struct iwl_trans *trans, + struct iwl_txq *txq) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct device *dev = trans->dev; + + /* De-alloc circular buffer of TFDs */ + if (txq->tfds) { + dma_free_coherent(dev, + trans_pcie->txqs.tfd.size * txq->n_window, + txq->tfds, txq->dma_addr); + dma_free_coherent(dev, + sizeof(*txq->first_tb_bufs) * txq->n_window, + txq->first_tb_bufs, txq->first_tb_dma); + } + + kfree(txq->entries); + if (txq->bc_tbl.addr) + dma_pool_free(trans_pcie->txqs.bc_pool, + txq->bc_tbl.addr, txq->bc_tbl.dma); + kfree(txq); +} + +/* + * iwl_pcie_txq_free - Deallocate DMA queue. + * @txq: Transmit queue to deallocate. + * + * Empty queue by removing and destroying all BD's. + * Free all buffers. + * 0-fill, but do not free "txq" descriptor structure. + */ +static void iwl_txq_gen2_free(struct iwl_trans *trans, int txq_id) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwl_txq *txq; + int i; + + if (WARN_ONCE(txq_id >= IWL_MAX_TVQM_QUEUES, + "queue %d out of range", txq_id)) + return; + + txq = trans_pcie->txqs.txq[txq_id]; + + if (WARN_ON(!txq)) + return; + + iwl_txq_gen2_unmap(trans, txq_id); + + /* De-alloc array of command/tx buffers */ + if (txq_id == trans_pcie->txqs.cmd.q_id) + for (i = 0; i < txq->n_window; i++) { + kfree_sensitive(txq->entries[i].cmd); + kfree_sensitive(txq->entries[i].free_buf); + } + del_timer_sync(&txq->stuck_timer); + + iwl_txq_gen2_free_memory(trans, txq); + + trans_pcie->txqs.txq[txq_id] = NULL; + + clear_bit(txq_id, trans_pcie->txqs.queue_used); +} + +static struct iwl_txq * +iwl_txq_dyn_alloc_dma(struct iwl_trans *trans, int size, unsigned int timeout) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + size_t bc_tbl_size, bc_tbl_entries; + struct iwl_txq *txq; + int ret; + + WARN_ON(!trans_pcie->txqs.bc_tbl_size); + + bc_tbl_size = trans_pcie->txqs.bc_tbl_size; + bc_tbl_entries = bc_tbl_size / sizeof(u16); + + if (WARN_ON(size > bc_tbl_entries)) + return ERR_PTR(-EINVAL); + + txq = kzalloc(sizeof(*txq), GFP_KERNEL); + if (!txq) + return ERR_PTR(-ENOMEM); + + txq->bc_tbl.addr = dma_pool_alloc(trans_pcie->txqs.bc_pool, GFP_KERNEL, + &txq->bc_tbl.dma); + if (!txq->bc_tbl.addr) { + IWL_ERR(trans, "Scheduler BC Table allocation failed\n"); + kfree(txq); + return ERR_PTR(-ENOMEM); + } + + ret = iwl_pcie_txq_alloc(trans, txq, size, false); + if (ret) { + IWL_ERR(trans, "Tx queue alloc failed\n"); + goto error; + } + ret = iwl_txq_init(trans, txq, size, false); + if (ret) { + IWL_ERR(trans, "Tx queue init failed\n"); + goto error; + } + + txq->wd_timeout = msecs_to_jiffies(timeout); + + return txq; + +error: + iwl_txq_gen2_free_memory(trans, txq); + return ERR_PTR(ret); +} + +static int iwl_pcie_txq_alloc_response(struct iwl_trans *trans, + struct iwl_txq *txq, + struct iwl_host_cmd *hcmd) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwl_tx_queue_cfg_rsp *rsp; + int ret, qid; + u32 wr_ptr; + + if (WARN_ON(iwl_rx_packet_payload_len(hcmd->resp_pkt) != + sizeof(*rsp))) { + ret = -EINVAL; + goto error_free_resp; + } + + rsp = (void *)hcmd->resp_pkt->data; + qid = le16_to_cpu(rsp->queue_number); + wr_ptr = le16_to_cpu(rsp->write_pointer); + + if (qid >= ARRAY_SIZE(trans_pcie->txqs.txq)) { + WARN_ONCE(1, "queue index %d unsupported", qid); + ret = -EIO; + goto error_free_resp; + } + + if (test_and_set_bit(qid, trans_pcie->txqs.queue_used)) { + WARN_ONCE(1, "queue %d already used", qid); + ret = -EIO; + goto error_free_resp; + } + + if (WARN_ONCE(trans_pcie->txqs.txq[qid], + "queue %d already allocated\n", qid)) { + ret = -EIO; + goto error_free_resp; + } + + txq->id = qid; + trans_pcie->txqs.txq[qid] = txq; + wr_ptr &= (trans->trans_cfg->base_params->max_tfd_queue_size - 1); + + /* Place first TFD at index corresponding to start sequence number */ + txq->read_ptr = wr_ptr; + txq->write_ptr = wr_ptr; + + IWL_DEBUG_TX_QUEUES(trans, "Activate queue %d\n", qid); + + iwl_free_resp(hcmd); + return qid; + +error_free_resp: + iwl_free_resp(hcmd); + iwl_txq_gen2_free_memory(trans, txq); + return ret; +} + +int iwl_txq_dyn_alloc(struct iwl_trans *trans, u32 flags, u32 sta_mask, + u8 tid, int size, unsigned int timeout) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwl_txq *txq; + union { + struct iwl_tx_queue_cfg_cmd old; + struct iwl_scd_queue_cfg_cmd new; + } cmd; + struct iwl_host_cmd hcmd = { + .flags = CMD_WANT_SKB, + }; + int ret; + + /* take the min with bytecount table entries allowed */ + size = min_t(u32, size, trans_pcie->txqs.bc_tbl_size / sizeof(u16)); + /* but must be power of 2 values for calculating read/write pointers */ + size = rounddown_pow_of_two(size); + + if (trans->trans_cfg->device_family == IWL_DEVICE_FAMILY_BZ && + trans->hw_rev_step == SILICON_A_STEP) { + size = 4096; + txq = iwl_txq_dyn_alloc_dma(trans, size, timeout); + } else { + do { + txq = iwl_txq_dyn_alloc_dma(trans, size, timeout); + if (!IS_ERR(txq)) + break; + + IWL_DEBUG_TX_QUEUES(trans, + "Failed allocating TXQ of size %d for sta mask %x tid %d, ret: %ld\n", + size, sta_mask, tid, + PTR_ERR(txq)); + size /= 2; + } while (size >= 16); + } + + if (IS_ERR(txq)) + return PTR_ERR(txq); + + if (trans_pcie->txqs.queue_alloc_cmd_ver == 0) { + memset(&cmd.old, 0, sizeof(cmd.old)); + cmd.old.tfdq_addr = cpu_to_le64(txq->dma_addr); + cmd.old.byte_cnt_addr = cpu_to_le64(txq->bc_tbl.dma); + cmd.old.cb_size = cpu_to_le32(TFD_QUEUE_CB_SIZE(size)); + cmd.old.flags = cpu_to_le16(flags | TX_QUEUE_CFG_ENABLE_QUEUE); + cmd.old.tid = tid; + + if (hweight32(sta_mask) != 1) { + ret = -EINVAL; + goto error; + } + cmd.old.sta_id = ffs(sta_mask) - 1; + + hcmd.id = SCD_QUEUE_CFG; + hcmd.len[0] = sizeof(cmd.old); + hcmd.data[0] = &cmd.old; + } else if (trans_pcie->txqs.queue_alloc_cmd_ver == 3) { + memset(&cmd.new, 0, sizeof(cmd.new)); + cmd.new.operation = cpu_to_le32(IWL_SCD_QUEUE_ADD); + cmd.new.u.add.tfdq_dram_addr = cpu_to_le64(txq->dma_addr); + cmd.new.u.add.bc_dram_addr = cpu_to_le64(txq->bc_tbl.dma); + cmd.new.u.add.cb_size = cpu_to_le32(TFD_QUEUE_CB_SIZE(size)); + cmd.new.u.add.flags = cpu_to_le32(flags); + cmd.new.u.add.sta_mask = cpu_to_le32(sta_mask); + cmd.new.u.add.tid = tid; + + hcmd.id = WIDE_ID(DATA_PATH_GROUP, SCD_QUEUE_CONFIG_CMD); + hcmd.len[0] = sizeof(cmd.new); + hcmd.data[0] = &cmd.new; + } else { + ret = -EOPNOTSUPP; + goto error; + } + + ret = iwl_trans_send_cmd(trans, &hcmd); + if (ret) + goto error; + + return iwl_pcie_txq_alloc_response(trans, txq, &hcmd); + +error: + iwl_txq_gen2_free_memory(trans, txq); + return ret; +} + +void iwl_txq_dyn_free(struct iwl_trans *trans, int queue) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + + if (WARN(queue >= IWL_MAX_TVQM_QUEUES, + "queue %d out of range", queue)) + return; + + /* + * Upon HW Rfkill - we stop the device, and then stop the queues + * in the op_mode. Just for the sake of the simplicity of the op_mode, + * allow the op_mode to call txq_disable after it already called + * stop_device. + */ + if (!test_and_clear_bit(queue, trans_pcie->txqs.queue_used)) { + WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status), + "queue %d not used", queue); + return; + } + + iwl_txq_gen2_free(trans, queue); + + IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", queue); +} + +void iwl_txq_gen2_tx_free(struct iwl_trans *trans) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + int i; + + memset(trans_pcie->txqs.queue_used, 0, + sizeof(trans_pcie->txqs.queue_used)); + + /* Free all TX queues */ + for (i = 0; i < ARRAY_SIZE(trans_pcie->txqs.txq); i++) { + if (!trans_pcie->txqs.txq[i]) + continue; + + iwl_txq_gen2_free(trans, i); + } +} + +int iwl_txq_gen2_init(struct iwl_trans *trans, int txq_id, int queue_size) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwl_txq *queue; + int ret; + + /* alloc and init the tx queue */ + if (!trans_pcie->txqs.txq[txq_id]) { + queue = kzalloc(sizeof(*queue), GFP_KERNEL); + if (!queue) { + IWL_ERR(trans, "Not enough memory for tx queue\n"); + return -ENOMEM; + } + trans_pcie->txqs.txq[txq_id] = queue; + ret = iwl_pcie_txq_alloc(trans, queue, queue_size, true); + if (ret) { + IWL_ERR(trans, "Tx %d queue init failed\n", txq_id); + goto error; + } + } else { + queue = trans_pcie->txqs.txq[txq_id]; + } + + ret = iwl_txq_init(trans, queue, queue_size, + (txq_id == trans_pcie->txqs.cmd.q_id)); + if (ret) { + IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id); + goto error; + } + trans_pcie->txqs.txq[txq_id]->id = txq_id; + set_bit(txq_id, trans_pcie->txqs.queue_used); + + return 0; + +error: + iwl_txq_gen2_tx_free(trans); + return ret; +} /*************** HOST COMMAND QUEUE FUNCTIONS *****/ @@ -28,7 +1204,7 @@ int iwl_pcie_gen2_enqueue_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd) { struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); - struct iwl_txq *txq = trans->txqs.txq[trans->txqs.cmd.q_id]; + struct iwl_txq *txq = trans_pcie->txqs.txq[trans_pcie->txqs.cmd.q_id]; struct iwl_device_cmd *out_cmd; struct iwl_cmd_meta *out_meta; void *dup_buf = NULL; @@ -130,7 +1306,7 @@ int iwl_pcie_gen2_enqueue_hcmd(struct iwl_trans *trans, out_cmd = txq->entries[idx].cmd; out_meta = &txq->entries[idx].meta; - /* re-initialize to NULL */ + /* re-initialize, this also marks the SG list as unused */ memset(out_meta, 0, sizeof(*out_meta)); if (cmd->flags & CMD_WANT_SKB) out_meta->source = cmd; @@ -143,7 +1319,7 @@ int iwl_pcie_gen2_enqueue_hcmd(struct iwl_trans *trans, cpu_to_le16(cmd_size - sizeof(struct iwl_cmd_header_wide)); out_cmd->hdr_wide.reserved = 0; out_cmd->hdr_wide.sequence = - cpu_to_le16(QUEUE_TO_SEQ(trans->txqs.cmd.q_id) | + cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->txqs.cmd.q_id) | INDEX_TO_SEQ(txq->write_ptr)); cmd_pos = sizeof(struct iwl_cmd_header_wide); @@ -191,7 +1367,7 @@ int iwl_pcie_gen2_enqueue_hcmd(struct iwl_trans *trans, "Sending command %s (%.2x.%.2x), seq: 0x%04X, %d bytes at %d[%d]:%d\n", iwl_get_cmd_string(trans, cmd->id), group_id, out_cmd->hdr.cmd, le16_to_cpu(out_cmd->hdr.sequence), - cmd_size, txq->write_ptr, idx, trans->txqs.cmd.q_id); + cmd_size, txq->write_ptr, idx, trans_pcie->txqs.cmd.q_id); /* start the TFD with the minimum copy bytes */ tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE); diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/tx.c b/drivers/net/wireless/intel/iwlwifi/pcie/tx.c index fa8eba47dc4c..9fe050f0ddc1 100644 --- a/drivers/net/wireless/intel/iwlwifi/pcie/tx.c +++ b/drivers/net/wireless/intel/iwlwifi/pcie/tx.c @@ -1,16 +1,22 @@ // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause /* - * Copyright (C) 2003-2014, 2018-2021, 2023 Intel Corporation + * Copyright (C) 2003-2014, 2018-2021, 2023-2024 Intel Corporation * Copyright (C) 2013-2015 Intel Mobile Communications GmbH * Copyright (C) 2016-2017 Intel Deutschland GmbH */ #include <linux/etherdevice.h> #include <linux/ieee80211.h> +#include <linux/dmapool.h> #include <linux/slab.h> #include <linux/sched.h> +#include <linux/tcp.h> #include <net/ip6_checksum.h> #include <net/tso.h> +#include "fw/api/commands.h" +#include "fw/api/datapath.h" +#include "fw/api/debug.h" +#include "iwl-fh.h" #include "iwl-debug.h" #include "iwl-csr.h" #include "iwl-prph.h" @@ -72,6 +78,7 @@ void iwl_pcie_free_dma_ptr(struct iwl_trans *trans, struct iwl_dma_ptr *ptr) static void iwl_pcie_txq_inc_wr_ptr(struct iwl_trans *trans, struct iwl_txq *txq) { + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); u32 reg = 0; int txq_id = txq->id; @@ -84,7 +91,7 @@ static void iwl_pcie_txq_inc_wr_ptr(struct iwl_trans *trans, * 3. there is a chance that the NIC is asleep */ if (!trans->trans_cfg->base_params->shadow_reg_enable && - txq_id != trans->txqs.cmd.q_id && + txq_id != trans_pcie->txqs.cmd.q_id && test_bit(STATUS_TPOWER_PMI, &trans->status)) { /* * wake up nic if it's powered down ... @@ -115,12 +122,13 @@ static void iwl_pcie_txq_inc_wr_ptr(struct iwl_trans *trans, void iwl_pcie_txq_check_wrptrs(struct iwl_trans *trans) { + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); int i; for (i = 0; i < trans->trans_cfg->base_params->num_of_queues; i++) { - struct iwl_txq *txq = trans->txqs.txq[i]; + struct iwl_txq *txq = trans_pcie->txqs.txq[i]; - if (!test_bit(i, trans->txqs.queue_used)) + if (!test_bit(i, trans_pcie->txqs.queue_used)) continue; spin_lock_bh(&txq->lock); @@ -132,23 +140,43 @@ void iwl_pcie_txq_check_wrptrs(struct iwl_trans *trans) } } +static inline void iwl_pcie_gen1_tfd_set_tb(struct iwl_tfd *tfd, + u8 idx, dma_addr_t addr, u16 len) +{ + struct iwl_tfd_tb *tb = &tfd->tbs[idx]; + u16 hi_n_len = len << 4; + + put_unaligned_le32(addr, &tb->lo); + hi_n_len |= iwl_get_dma_hi_addr(addr); + + tb->hi_n_len = cpu_to_le16(hi_n_len); + + tfd->num_tbs = idx + 1; +} + +static inline u8 iwl_txq_gen1_tfd_get_num_tbs(struct iwl_tfd *tfd) +{ + return tfd->num_tbs & 0x1f; +} + static int iwl_pcie_txq_build_tfd(struct iwl_trans *trans, struct iwl_txq *txq, dma_addr_t addr, u16 len, bool reset) { + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); void *tfd; u32 num_tbs; - tfd = (u8 *)txq->tfds + trans->txqs.tfd.size * txq->write_ptr; + tfd = (u8 *)txq->tfds + trans_pcie->txqs.tfd.size * txq->write_ptr; if (reset) - memset(tfd, 0, trans->txqs.tfd.size); + memset(tfd, 0, trans_pcie->txqs.tfd.size); - num_tbs = iwl_txq_gen1_tfd_get_num_tbs(trans, tfd); + num_tbs = iwl_txq_gen1_tfd_get_num_tbs(tfd); /* Each TFD can point to a maximum max_tbs Tx buffers */ - if (num_tbs >= trans->txqs.tfd.max_tbs) { + if (num_tbs >= trans_pcie->txqs.tfd.max_tbs) { IWL_ERR(trans, "Error can not send more than %d chunks\n", - trans->txqs.tfd.max_tbs); + trans_pcie->txqs.tfd.max_tbs); return -EINVAL; } @@ -156,7 +184,7 @@ static int iwl_pcie_txq_build_tfd(struct iwl_trans *trans, struct iwl_txq *txq, "Unaligned address = %llx\n", (unsigned long long)addr)) return -EINVAL; - iwl_pcie_gen1_tfd_set_tb(trans, tfd, num_tbs, addr, len); + iwl_pcie_gen1_tfd_set_tb(tfd, num_tbs, addr, len); return num_tbs; } @@ -181,36 +209,206 @@ static void iwl_pcie_clear_cmd_in_flight(struct iwl_trans *trans) spin_unlock(&trans_pcie->reg_lock); } +static void iwl_pcie_free_and_unmap_tso_page(struct iwl_trans *trans, + struct page *page) +{ + struct iwl_tso_page_info *info = IWL_TSO_PAGE_INFO(page_address(page)); + + /* Decrease internal use count and unmap/free page if needed */ + if (refcount_dec_and_test(&info->use_count)) { + dma_unmap_page(trans->dev, info->dma_addr, PAGE_SIZE, + DMA_TO_DEVICE); + + __free_page(page); + } +} + +void iwl_pcie_free_tso_pages(struct iwl_trans *trans, struct sk_buff *skb, + struct iwl_cmd_meta *cmd_meta) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct page **page_ptr; + struct page *next; + + page_ptr = (void *)((u8 *)skb->cb + trans_pcie->txqs.page_offs); + next = *page_ptr; + *page_ptr = NULL; + + while (next) { + struct iwl_tso_page_info *info; + struct page *tmp = next; + + info = IWL_TSO_PAGE_INFO(page_address(next)); + next = info->next; + + /* Unmap the scatter gather list that is on the last page */ + if (!next && cmd_meta->sg_offset) { + struct sg_table *sgt; + + sgt = (void *)((u8 *)page_address(tmp) + + cmd_meta->sg_offset); + + dma_unmap_sgtable(trans->dev, sgt, DMA_TO_DEVICE, 0); + } + + iwl_pcie_free_and_unmap_tso_page(trans, tmp); + } +} + +static inline dma_addr_t +iwl_txq_gen1_tfd_tb_get_addr(struct iwl_tfd *tfd, u8 idx) +{ + struct iwl_tfd_tb *tb = &tfd->tbs[idx]; + dma_addr_t addr; + dma_addr_t hi_len; + + addr = get_unaligned_le32(&tb->lo); + + if (sizeof(dma_addr_t) <= sizeof(u32)) + return addr; + + hi_len = le16_to_cpu(tb->hi_n_len) & 0xF; + + /* + * shift by 16 twice to avoid warnings on 32-bit + * (where this code never runs anyway due to the + * if statement above) + */ + return addr | ((hi_len << 16) << 16); +} + +static void iwl_txq_set_tfd_invalid_gen1(struct iwl_trans *trans, + struct iwl_tfd *tfd) +{ + tfd->num_tbs = 0; + + iwl_pcie_gen1_tfd_set_tb(tfd, 0, trans->invalid_tx_cmd.dma, + trans->invalid_tx_cmd.size); +} + +static void iwl_txq_gen1_tfd_unmap(struct iwl_trans *trans, + struct iwl_cmd_meta *meta, + struct iwl_txq *txq, int index) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + int i, num_tbs; + struct iwl_tfd *tfd = iwl_txq_get_tfd(trans, txq, index); + + /* Sanity check on number of chunks */ + num_tbs = iwl_txq_gen1_tfd_get_num_tbs(tfd); + + if (num_tbs > trans_pcie->txqs.tfd.max_tbs) { + IWL_ERR(trans, "Too many chunks: %i\n", num_tbs); + /* @todo issue fatal error, it is quite serious situation */ + return; + } + + /* TB1 is mapped directly, the rest is the TSO page and SG list. */ + if (meta->sg_offset) + num_tbs = 2; + + /* first TB is never freed - it's the bidirectional DMA data */ + + for (i = 1; i < num_tbs; i++) { + if (meta->tbs & BIT(i)) + dma_unmap_page(trans->dev, + iwl_txq_gen1_tfd_tb_get_addr(tfd, i), + iwl_txq_gen1_tfd_tb_get_len(trans, + tfd, i), + DMA_TO_DEVICE); + else + dma_unmap_single(trans->dev, + iwl_txq_gen1_tfd_tb_get_addr(tfd, i), + iwl_txq_gen1_tfd_tb_get_len(trans, + tfd, i), + DMA_TO_DEVICE); + } + + meta->tbs = 0; + + iwl_txq_set_tfd_invalid_gen1(trans, tfd); +} + +/** + * iwl_txq_free_tfd - Free all chunks referenced by TFD [txq->q.read_ptr] + * @trans: transport private data + * @txq: tx queue + * @read_ptr: the TXQ read_ptr to free + * + * Does NOT advance any TFD circular buffer read/write indexes + * Does NOT free the TFD itself (which is within circular buffer) + */ +static void iwl_txq_free_tfd(struct iwl_trans *trans, struct iwl_txq *txq, + int read_ptr) +{ + /* rd_ptr is bounded by TFD_QUEUE_SIZE_MAX and + * idx is bounded by n_window + */ + int idx = iwl_txq_get_cmd_index(txq, read_ptr); + struct sk_buff *skb; + + lockdep_assert_held(&txq->reclaim_lock); + + if (!txq->entries) + return; + + /* We have only q->n_window txq->entries, but we use + * TFD_QUEUE_SIZE_MAX tfds + */ + if (trans->trans_cfg->gen2) + iwl_txq_gen2_tfd_unmap(trans, &txq->entries[idx].meta, + iwl_txq_get_tfd(trans, txq, read_ptr)); + else + iwl_txq_gen1_tfd_unmap(trans, &txq->entries[idx].meta, + txq, read_ptr); + + /* free SKB */ + skb = txq->entries[idx].skb; + + /* Can be called from irqs-disabled context + * If skb is not NULL, it means that the whole queue is being + * freed and that the queue is not empty - free the skb + */ + if (skb) { + iwl_op_mode_free_skb(trans->op_mode, skb); + txq->entries[idx].skb = NULL; + } +} + /* * iwl_pcie_txq_unmap - Unmap any remaining DMA mappings and free skb's */ static void iwl_pcie_txq_unmap(struct iwl_trans *trans, int txq_id) { - struct iwl_txq *txq = trans->txqs.txq[txq_id]; + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; if (!txq) { IWL_ERR(trans, "Trying to free a queue that wasn't allocated?\n"); return; } - spin_lock_bh(&txq->lock); + spin_lock_bh(&txq->reclaim_lock); + spin_lock(&txq->lock); while (txq->write_ptr != txq->read_ptr) { IWL_DEBUG_TX_REPLY(trans, "Q %d Free %d\n", txq_id, txq->read_ptr); - if (txq_id != trans->txqs.cmd.q_id) { + if (txq_id != trans_pcie->txqs.cmd.q_id) { struct sk_buff *skb = txq->entries[txq->read_ptr].skb; + struct iwl_cmd_meta *cmd_meta = + &txq->entries[txq->read_ptr].meta; if (WARN_ON_ONCE(!skb)) continue; - iwl_txq_free_tso_page(trans, skb); + iwl_pcie_free_tso_pages(trans, skb, cmd_meta); } - iwl_txq_free_tfd(trans, txq); + iwl_txq_free_tfd(trans, txq, txq->read_ptr); txq->read_ptr = iwl_txq_inc_wrap(trans, txq->read_ptr); if (txq->read_ptr == txq->write_ptr && - txq_id == trans->txqs.cmd.q_id) + txq_id == trans_pcie->txqs.cmd.q_id) iwl_pcie_clear_cmd_in_flight(trans); } @@ -220,10 +418,11 @@ static void iwl_pcie_txq_unmap(struct iwl_trans *trans, int txq_id) iwl_op_mode_free_skb(trans->op_mode, skb); } - spin_unlock_bh(&txq->lock); + spin_unlock(&txq->lock); + spin_unlock_bh(&txq->reclaim_lock); /* just in case - this queue may have been stopped */ - iwl_wake_queue(trans, txq); + iwl_trans_pcie_wake_queue(trans, txq); } /* @@ -236,7 +435,8 @@ static void iwl_pcie_txq_unmap(struct iwl_trans *trans, int txq_id) */ static void iwl_pcie_txq_free(struct iwl_trans *trans, int txq_id) { - struct iwl_txq *txq = trans->txqs.txq[txq_id]; + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; struct device *dev = trans->dev; int i; @@ -246,7 +446,7 @@ static void iwl_pcie_txq_free(struct iwl_trans *trans, int txq_id) iwl_pcie_txq_unmap(trans, txq_id); /* De-alloc array of command/tx buffers */ - if (txq_id == trans->txqs.cmd.q_id) + if (txq_id == trans_pcie->txqs.cmd.q_id) for (i = 0; i < txq->n_window; i++) { kfree_sensitive(txq->entries[i].cmd); kfree_sensitive(txq->entries[i].free_buf); @@ -255,7 +455,7 @@ static void iwl_pcie_txq_free(struct iwl_trans *trans, int txq_id) /* De-alloc circular buffer of TFDs */ if (txq->tfds) { dma_free_coherent(dev, - trans->txqs.tfd.size * + trans_pcie->txqs.tfd.size * trans->trans_cfg->base_params->max_tfd_queue_size, txq->tfds, txq->dma_addr); txq->dma_addr = 0; @@ -285,9 +485,10 @@ void iwl_pcie_tx_start(struct iwl_trans *trans, u32 scd_base_addr) SCD_CONTEXT_MEM_LOWER_BOUND) / sizeof(u32); /* make sure all queue are not stopped/used */ - memset(trans->txqs.queue_stopped, 0, - sizeof(trans->txqs.queue_stopped)); - memset(trans->txqs.queue_used, 0, sizeof(trans->txqs.queue_used)); + memset(trans_pcie->txqs.queue_stopped, 0, + sizeof(trans_pcie->txqs.queue_stopped)); + memset(trans_pcie->txqs.queue_used, 0, + sizeof(trans_pcie->txqs.queue_used)); trans_pcie->scd_base_addr = iwl_read_prph(trans, SCD_SRAM_BASE_ADDR); @@ -301,7 +502,7 @@ void iwl_pcie_tx_start(struct iwl_trans *trans, u32 scd_base_addr) NULL, clear_dwords); iwl_write_prph(trans, SCD_DRAM_BASE_ADDR, - trans->txqs.scd_bc_tbls.dma >> 10); + trans_pcie->txqs.scd_bc_tbls.dma >> 10); /* The chain extension of the SCD doesn't work well. This feature is * enabled by default by the HW, so we need to disable it manually. @@ -309,9 +510,9 @@ void iwl_pcie_tx_start(struct iwl_trans *trans, u32 scd_base_addr) if (trans->trans_cfg->base_params->scd_chain_ext_wa) iwl_write_prph(trans, SCD_CHAINEXT_EN, 0); - iwl_trans_ac_txq_enable(trans, trans->txqs.cmd.q_id, - trans->txqs.cmd.fifo, - trans->txqs.cmd.wdg_timeout); + iwl_trans_ac_txq_enable(trans, trans_pcie->txqs.cmd.q_id, + trans_pcie->txqs.cmd.fifo, + trans_pcie->txqs.cmd.wdg_timeout); /* Activate all Tx DMA/FIFO channels */ iwl_scd_activate_fifos(trans); @@ -347,7 +548,7 @@ void iwl_trans_pcie_tx_reset(struct iwl_trans *trans) for (txq_id = 0; txq_id < trans->trans_cfg->base_params->num_of_queues; txq_id++) { - struct iwl_txq *txq = trans->txqs.txq[txq_id]; + struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; if (trans->trans_cfg->gen2) iwl_write_direct64(trans, FH_MEM_CBBC_QUEUE(trans, txq_id), @@ -422,9 +623,10 @@ int iwl_pcie_tx_stop(struct iwl_trans *trans) * queues. This happens when we have an rfkill interrupt. * Since we stop Tx altogether - mark the queues as stopped. */ - memset(trans->txqs.queue_stopped, 0, - sizeof(trans->txqs.queue_stopped)); - memset(trans->txqs.queue_used, 0, sizeof(trans->txqs.queue_used)); + memset(trans_pcie->txqs.queue_stopped, 0, + sizeof(trans_pcie->txqs.queue_stopped)); + memset(trans_pcie->txqs.queue_used, 0, + sizeof(trans_pcie->txqs.queue_used)); /* This can happen: start_hw, stop_device */ if (!trans_pcie->txq_memory) @@ -448,7 +650,8 @@ void iwl_pcie_tx_free(struct iwl_trans *trans) int txq_id; struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); - memset(trans->txqs.queue_used, 0, sizeof(trans->txqs.queue_used)); + memset(trans_pcie->txqs.queue_used, 0, + sizeof(trans_pcie->txqs.queue_used)); /* Tx queues */ if (trans_pcie->txq_memory) { @@ -456,7 +659,7 @@ void iwl_pcie_tx_free(struct iwl_trans *trans) txq_id < trans->trans_cfg->base_params->num_of_queues; txq_id++) { iwl_pcie_txq_free(trans, txq_id); - trans->txqs.txq[txq_id] = NULL; + trans_pcie->txqs.txq[txq_id] = NULL; } } @@ -465,7 +668,135 @@ void iwl_pcie_tx_free(struct iwl_trans *trans) iwl_pcie_free_dma_ptr(trans, &trans_pcie->kw); - iwl_pcie_free_dma_ptr(trans, &trans->txqs.scd_bc_tbls); + iwl_pcie_free_dma_ptr(trans, &trans_pcie->txqs.scd_bc_tbls); +} + +void iwl_txq_log_scd_error(struct iwl_trans *trans, struct iwl_txq *txq) +{ + u32 txq_id = txq->id; + u32 status; + bool active; + u8 fifo; + + if (trans->trans_cfg->gen2) { + IWL_ERR(trans, "Queue %d is stuck %d %d\n", txq_id, + txq->read_ptr, txq->write_ptr); + /* TODO: access new SCD registers and dump them */ + return; + } + + status = iwl_read_prph(trans, SCD_QUEUE_STATUS_BITS(txq_id)); + fifo = (status >> SCD_QUEUE_STTS_REG_POS_TXF) & 0x7; + active = !!(status & BIT(SCD_QUEUE_STTS_REG_POS_ACTIVE)); + + IWL_ERR(trans, + "Queue %d is %sactive on fifo %d and stuck for %u ms. SW [%d, %d] HW [%d, %d] FH TRB=0x0%x\n", + txq_id, active ? "" : "in", fifo, + jiffies_to_msecs(txq->wd_timeout), + txq->read_ptr, txq->write_ptr, + iwl_read_prph(trans, SCD_QUEUE_RDPTR(txq_id)) & + (trans->trans_cfg->base_params->max_tfd_queue_size - 1), + iwl_read_prph(trans, SCD_QUEUE_WRPTR(txq_id)) & + (trans->trans_cfg->base_params->max_tfd_queue_size - 1), + iwl_read_direct32(trans, FH_TX_TRB_REG(fifo))); +} + +static void iwl_txq_stuck_timer(struct timer_list *t) +{ + struct iwl_txq *txq = from_timer(txq, t, stuck_timer); + struct iwl_trans *trans = txq->trans; + + spin_lock(&txq->lock); + /* check if triggered erroneously */ + if (txq->read_ptr == txq->write_ptr) { + spin_unlock(&txq->lock); + return; + } + spin_unlock(&txq->lock); + + iwl_txq_log_scd_error(trans, txq); + + iwl_force_nmi(trans); +} + +int iwl_pcie_txq_alloc(struct iwl_trans *trans, struct iwl_txq *txq, + int slots_num, bool cmd_queue) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + size_t num_entries = trans->trans_cfg->gen2 ? + slots_num : trans->trans_cfg->base_params->max_tfd_queue_size; + size_t tfd_sz; + size_t tb0_buf_sz; + int i; + + if (WARN_ONCE(slots_num <= 0, "Invalid slots num:%d\n", slots_num)) + return -EINVAL; + + if (WARN_ON(txq->entries || txq->tfds)) + return -EINVAL; + + tfd_sz = trans_pcie->txqs.tfd.size * num_entries; + + timer_setup(&txq->stuck_timer, iwl_txq_stuck_timer, 0); + txq->trans = trans; + + txq->n_window = slots_num; + + txq->entries = kcalloc(slots_num, + sizeof(struct iwl_pcie_txq_entry), + GFP_KERNEL); + + if (!txq->entries) + goto error; + + if (cmd_queue) + for (i = 0; i < slots_num; i++) { + txq->entries[i].cmd = + kmalloc(sizeof(struct iwl_device_cmd), + GFP_KERNEL); + if (!txq->entries[i].cmd) + goto error; + } + + /* Circular buffer of transmit frame descriptors (TFDs), + * shared with device + */ + txq->tfds = dma_alloc_coherent(trans->dev, tfd_sz, + &txq->dma_addr, GFP_KERNEL); + if (!txq->tfds) + goto error; + + BUILD_BUG_ON(sizeof(*txq->first_tb_bufs) != IWL_FIRST_TB_SIZE_ALIGN); + + tb0_buf_sz = sizeof(*txq->first_tb_bufs) * slots_num; + + txq->first_tb_bufs = dma_alloc_coherent(trans->dev, tb0_buf_sz, + &txq->first_tb_dma, + GFP_KERNEL); + if (!txq->first_tb_bufs) + goto err_free_tfds; + + for (i = 0; i < num_entries; i++) { + void *tfd = iwl_txq_get_tfd(trans, txq, i); + + if (trans->trans_cfg->gen2) + iwl_txq_set_tfd_invalid_gen2(trans, tfd); + else + iwl_txq_set_tfd_invalid_gen1(trans, tfd); + } + + return 0; +err_free_tfds: + dma_free_coherent(trans->dev, tfd_sz, txq->tfds, txq->dma_addr); + txq->tfds = NULL; +error: + if (txq->entries && cmd_queue) + for (i = 0; i < slots_num; i++) + kfree(txq->entries[i].cmd); + kfree(txq->entries); + txq->entries = NULL; + + return -ENOMEM; } /* @@ -491,7 +822,7 @@ static int iwl_pcie_tx_alloc(struct iwl_trans *trans) goto error; } - ret = iwl_pcie_alloc_dma_ptr(trans, &trans->txqs.scd_bc_tbls, + ret = iwl_pcie_alloc_dma_ptr(trans, &trans_pcie->txqs.scd_bc_tbls, bc_tbls_size); if (ret) { IWL_ERR(trans, "Scheduler BC Table allocation failed\n"); @@ -517,7 +848,7 @@ static int iwl_pcie_tx_alloc(struct iwl_trans *trans) /* Alloc and init all Tx queues, including the command queue (#4/#9) */ for (txq_id = 0; txq_id < trans->trans_cfg->base_params->num_of_queues; txq_id++) { - bool cmd_queue = (txq_id == trans->txqs.cmd.q_id); + bool cmd_queue = (txq_id == trans_pcie->txqs.cmd.q_id); if (cmd_queue) slots_num = max_t(u32, IWL_CMD_QUEUE_SIZE, @@ -525,14 +856,14 @@ static int iwl_pcie_tx_alloc(struct iwl_trans *trans) else slots_num = max_t(u32, IWL_DEFAULT_QUEUE_SIZE, trans->cfg->min_ba_txq_size); - trans->txqs.txq[txq_id] = &trans_pcie->txq_memory[txq_id]; - ret = iwl_txq_alloc(trans, trans->txqs.txq[txq_id], slots_num, - cmd_queue); + trans_pcie->txqs.txq[txq_id] = &trans_pcie->txq_memory[txq_id]; + ret = iwl_pcie_txq_alloc(trans, trans_pcie->txqs.txq[txq_id], + slots_num, cmd_queue); if (ret) { IWL_ERR(trans, "Tx %d queue alloc failed\n", txq_id); goto error; } - trans->txqs.txq[txq_id]->id = txq_id; + trans_pcie->txqs.txq[txq_id]->id = txq_id; } return 0; @@ -543,6 +874,69 @@ error: return ret; } +/* + * iwl_queue_init - Initialize queue's high/low-water and read/write indexes + */ +static int iwl_queue_init(struct iwl_txq *q, int slots_num) +{ + q->n_window = slots_num; + + /* slots_num must be power-of-two size, otherwise + * iwl_txq_get_cmd_index is broken. + */ + if (WARN_ON(!is_power_of_2(slots_num))) + return -EINVAL; + + q->low_mark = q->n_window / 4; + if (q->low_mark < 4) + q->low_mark = 4; + + q->high_mark = q->n_window / 8; + if (q->high_mark < 2) + q->high_mark = 2; + + q->write_ptr = 0; + q->read_ptr = 0; + + return 0; +} + +int iwl_txq_init(struct iwl_trans *trans, struct iwl_txq *txq, + int slots_num, bool cmd_queue) +{ + u32 tfd_queue_max_size = + trans->trans_cfg->base_params->max_tfd_queue_size; + int ret; + + txq->need_update = false; + + /* max_tfd_queue_size must be power-of-two size, otherwise + * iwl_txq_inc_wrap and iwl_txq_dec_wrap are broken. + */ + if (WARN_ONCE(tfd_queue_max_size & (tfd_queue_max_size - 1), + "Max tfd queue size must be a power of two, but is %d", + tfd_queue_max_size)) + return -EINVAL; + + /* Initialize queue's high/low-water marks, and head/tail indexes */ + ret = iwl_queue_init(txq, slots_num); + if (ret) + return ret; + + spin_lock_init(&txq->lock); + spin_lock_init(&txq->reclaim_lock); + + if (cmd_queue) { + static struct lock_class_key iwl_txq_cmd_queue_lock_class; + + lockdep_set_class(&txq->lock, &iwl_txq_cmd_queue_lock_class); + } + + __skb_queue_head_init(&txq->overflow_q); + + return 0; +} + int iwl_pcie_tx_init(struct iwl_trans *trans) { struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); @@ -571,7 +965,7 @@ int iwl_pcie_tx_init(struct iwl_trans *trans) /* Alloc and init all Tx queues, including the command queue (#4/#9) */ for (txq_id = 0; txq_id < trans->trans_cfg->base_params->num_of_queues; txq_id++) { - bool cmd_queue = (txq_id == trans->txqs.cmd.q_id); + bool cmd_queue = (txq_id == trans_pcie->txqs.cmd.q_id); if (cmd_queue) slots_num = max_t(u32, IWL_CMD_QUEUE_SIZE, @@ -579,7 +973,7 @@ int iwl_pcie_tx_init(struct iwl_trans *trans) else slots_num = max_t(u32, IWL_DEFAULT_QUEUE_SIZE, trans->cfg->min_ba_txq_size); - ret = iwl_txq_init(trans, trans->txqs.txq[txq_id], slots_num, + ret = iwl_txq_init(trans, trans_pcie->txqs.txq[txq_id], slots_num, cmd_queue); if (ret) { IWL_ERR(trans, "Tx %d queue init failed\n", txq_id); @@ -593,7 +987,7 @@ int iwl_pcie_tx_init(struct iwl_trans *trans) * Circular buffer (TFD queue in DRAM) physical base address */ iwl_write_direct32(trans, FH_MEM_CBBC_QUEUE(trans, txq_id), - trans->txqs.txq[txq_id]->dma_addr >> 8); + trans_pcie->txqs.txq[txq_id]->dma_addr >> 8); } iwl_set_bits_prph(trans, SCD_GP_CTRL, SCD_GP_CTRL_AUTO_ACTIVE_MODE); @@ -641,6 +1035,42 @@ static int iwl_pcie_set_cmd_in_flight(struct iwl_trans *trans, return 0; } +static void iwl_txq_progress(struct iwl_txq *txq) +{ + lockdep_assert_held(&txq->lock); + + if (!txq->wd_timeout) + return; + + /* + * station is asleep and we send data - that must + * be uAPSD or PS-Poll. Don't rearm the timer. + */ + if (txq->frozen) + return; + + /* + * if empty delete timer, otherwise move timer forward + * since we're making progress on this queue + */ + if (txq->read_ptr == txq->write_ptr) + del_timer(&txq->stuck_timer); + else + mod_timer(&txq->stuck_timer, jiffies + txq->wd_timeout); +} + +static inline bool iwl_txq_used(const struct iwl_txq *q, int i, + int read_ptr, int write_ptr) +{ + int index = iwl_txq_get_cmd_index(q, i); + int r = iwl_txq_get_cmd_index(q, read_ptr); + int w = iwl_txq_get_cmd_index(q, write_ptr); + + return w >= r ? + (index >= r && index < w) : + !(index < r && index >= w); +} + /* * iwl_pcie_cmdq_reclaim - Reclaim TX command queue entries already Tx'd * @@ -650,7 +1080,8 @@ static int iwl_pcie_set_cmd_in_flight(struct iwl_trans *trans, */ static void iwl_pcie_cmdq_reclaim(struct iwl_trans *trans, int txq_id, int idx) { - struct iwl_txq *txq = trans->txqs.txq[txq_id]; + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; int nfreed = 0; u16 r; @@ -660,8 +1091,8 @@ static void iwl_pcie_cmdq_reclaim(struct iwl_trans *trans, int txq_id, int idx) r = iwl_txq_get_cmd_index(txq, txq->read_ptr); if (idx >= trans->trans_cfg->base_params->max_tfd_queue_size || - (!iwl_txq_used(txq, idx))) { - WARN_ONCE(test_bit(txq_id, trans->txqs.queue_used), + (!iwl_txq_used(txq, idx, txq->read_ptr, txq->write_ptr))) { + WARN_ONCE(test_bit(txq_id, trans_pcie->txqs.queue_used), "%s: Read index for DMA queue txq id (%d), index %d is out of range [0-%d] %d %d.\n", __func__, txq_id, idx, trans->trans_cfg->base_params->max_tfd_queue_size, @@ -720,11 +1151,11 @@ bool iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, u16 ssn, unsigned int wdg_timeout) { struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); - struct iwl_txq *txq = trans->txqs.txq[txq_id]; + struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; int fifo = -1; bool scd_bug = false; - if (test_and_set_bit(txq_id, trans->txqs.queue_used)) + if (test_and_set_bit(txq_id, trans_pcie->txqs.queue_used)) WARN_ONCE(1, "queue %d already used - expect issues", txq_id); txq->wd_timeout = msecs_to_jiffies(wdg_timeout); @@ -733,7 +1164,7 @@ bool iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, u16 ssn, fifo = cfg->fifo; /* Disable the scheduler prior configuring the cmd queue */ - if (txq_id == trans->txqs.cmd.q_id && + if (txq_id == trans_pcie->txqs.cmd.q_id && trans_pcie->scd_set_active) iwl_scd_enable_set_active(trans, 0); @@ -741,7 +1172,7 @@ bool iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, u16 ssn, iwl_scd_txq_set_inactive(trans, txq_id); /* Set this queue as a chain-building queue unless it is CMD */ - if (txq_id != trans->txqs.cmd.q_id) + if (txq_id != trans_pcie->txqs.cmd.q_id) iwl_scd_txq_set_chain(trans, txq_id); if (cfg->aggregate) { @@ -811,7 +1242,7 @@ bool iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, u16 ssn, SCD_QUEUE_STTS_REG_MSK); /* enable the scheduler for this queue (only) */ - if (txq_id == trans->txqs.cmd.q_id && + if (txq_id == trans_pcie->txqs.cmd.q_id && trans_pcie->scd_set_active) iwl_scd_enable_set_active(trans, BIT(txq_id)); @@ -830,7 +1261,8 @@ bool iwl_trans_pcie_txq_enable(struct iwl_trans *trans, int txq_id, u16 ssn, void iwl_trans_pcie_txq_set_shared_mode(struct iwl_trans *trans, u32 txq_id, bool shared_mode) { - struct iwl_txq *txq = trans->txqs.txq[txq_id]; + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; txq->ampdu = !shared_mode; } @@ -843,8 +1275,8 @@ void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id, SCD_TX_STTS_QUEUE_OFFSET(txq_id); static const u32 zero_val[4] = {}; - trans->txqs.txq[txq_id]->frozen_expiry_remainder = 0; - trans->txqs.txq[txq_id]->frozen = false; + trans_pcie->txqs.txq[txq_id]->frozen_expiry_remainder = 0; + trans_pcie->txqs.txq[txq_id]->frozen = false; /* * Upon HW Rfkill - we stop the device, and then stop the queues @@ -852,7 +1284,7 @@ void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id, * allow the op_mode to call txq_disable after it already called * stop_device. */ - if (!test_and_clear_bit(txq_id, trans->txqs.queue_used)) { + if (!test_and_clear_bit(txq_id, trans_pcie->txqs.queue_used)) { WARN_ONCE(test_bit(STATUS_DEVICE_ENABLED, &trans->status), "queue %d not used", txq_id); return; @@ -866,7 +1298,7 @@ void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id, } iwl_pcie_txq_unmap(trans, txq_id); - trans->txqs.txq[txq_id]->ampdu = false; + trans_pcie->txqs.txq[txq_id]->ampdu = false; IWL_DEBUG_TX_QUEUES(trans, "Deactivate queue %d\n", txq_id); } @@ -875,12 +1307,13 @@ void iwl_trans_pcie_txq_disable(struct iwl_trans *trans, int txq_id, static void iwl_trans_pcie_block_txq_ptrs(struct iwl_trans *trans, bool block) { + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); int i; for (i = 0; i < trans->trans_cfg->base_params->num_of_queues; i++) { - struct iwl_txq *txq = trans->txqs.txq[i]; + struct iwl_txq *txq = trans_pcie->txqs.txq[i]; - if (i == trans->txqs.cmd.q_id) + if (i == trans_pcie->txqs.cmd.q_id) continue; /* we skip the command queue (obviously) so it's OK to nest */ @@ -912,7 +1345,8 @@ static void iwl_trans_pcie_block_txq_ptrs(struct iwl_trans *trans, bool block) int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd) { - struct iwl_txq *txq = trans->txqs.txq[trans->txqs.cmd.q_id]; + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwl_txq *txq = trans_pcie->txqs.txq[trans_pcie->txqs.cmd.q_id]; struct iwl_device_cmd *out_cmd; struct iwl_cmd_meta *out_meta; void *dup_buf = NULL; @@ -1024,7 +1458,8 @@ int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans, out_cmd = txq->entries[idx].cmd; out_meta = &txq->entries[idx].meta; - memset(out_meta, 0, sizeof(*out_meta)); /* re-initialize to NULL */ + /* re-initialize, this also marks the SG list as unused */ + memset(out_meta, 0, sizeof(*out_meta)); if (cmd->flags & CMD_WANT_SKB) out_meta->source = cmd; @@ -1038,7 +1473,7 @@ int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans, sizeof(struct iwl_cmd_header_wide)); out_cmd->hdr_wide.reserved = 0; out_cmd->hdr_wide.sequence = - cpu_to_le16(QUEUE_TO_SEQ(trans->txqs.cmd.q_id) | + cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->txqs.cmd.q_id) | INDEX_TO_SEQ(txq->write_ptr)); cmd_pos = sizeof(struct iwl_cmd_header_wide); @@ -1046,7 +1481,7 @@ int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans, } else { out_cmd->hdr.cmd = iwl_cmd_opcode(cmd->id); out_cmd->hdr.sequence = - cpu_to_le16(QUEUE_TO_SEQ(trans->txqs.cmd.q_id) | + cpu_to_le16(QUEUE_TO_SEQ(trans_pcie->txqs.cmd.q_id) | INDEX_TO_SEQ(txq->write_ptr)); out_cmd->hdr.group_id = 0; @@ -1097,7 +1532,7 @@ int iwl_pcie_enqueue_hcmd(struct iwl_trans *trans, iwl_get_cmd_string(trans, cmd->id), group_id, out_cmd->hdr.cmd, le16_to_cpu(out_cmd->hdr.sequence), - cmd_size, txq->write_ptr, idx, trans->txqs.cmd.q_id); + cmd_size, txq->write_ptr, idx, trans_pcie->txqs.cmd.q_id); /* start the TFD with the minimum copy bytes */ tb0_size = min_t(int, copy_size, IWL_FIRST_TB_SIZE); @@ -1196,14 +1631,14 @@ void iwl_pcie_hcmd_complete(struct iwl_trans *trans, struct iwl_device_cmd *cmd; struct iwl_cmd_meta *meta; struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); - struct iwl_txq *txq = trans->txqs.txq[trans->txqs.cmd.q_id]; + struct iwl_txq *txq = trans_pcie->txqs.txq[trans_pcie->txqs.cmd.q_id]; /* If a Tx command is being handled and it isn't in the actual * command queue then there a command routing bug has been introduced * in the queue management code. */ - if (WARN(txq_id != trans->txqs.cmd.q_id, + if (WARN(txq_id != trans_pcie->txqs.cmd.q_id, "wrong command queue %d (should be %d), sequence 0x%X readp=%d writep=%d\n", - txq_id, trans->txqs.cmd.q_id, sequence, txq->read_ptr, + txq_id, trans_pcie->txqs.cmd.q_id, sequence, txq->read_ptr, txq->write_ptr)) { iwl_print_hex_error(trans, pkt, 32); return; @@ -1306,19 +1741,180 @@ static int iwl_fill_data_tbs(struct iwl_trans *trans, struct sk_buff *skb, } #ifdef CONFIG_INET +static void *iwl_pcie_get_page_hdr(struct iwl_trans *trans, + size_t len, struct sk_buff *skb) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwl_tso_hdr_page *p = this_cpu_ptr(trans_pcie->txqs.tso_hdr_page); + struct iwl_tso_page_info *info; + struct page **page_ptr; + dma_addr_t phys; + void *ret; + + page_ptr = (void *)((u8 *)skb->cb + trans_pcie->txqs.page_offs); + + if (WARN_ON(*page_ptr)) + return NULL; + + if (!p->page) + goto alloc; + + /* + * Check if there's enough room on this page + * + * Note that we put a page chaining pointer *last* in the + * page - we need it somewhere, and if it's there then we + * avoid DMA mapping the last bits of the page which may + * trigger the 32-bit boundary hardware bug. + * + * (see also get_workaround_page() in tx-gen2.c) + */ + if (((unsigned long)p->pos & ~PAGE_MASK) + len < IWL_TSO_PAGE_DATA_SIZE) { + info = IWL_TSO_PAGE_INFO(page_address(p->page)); + goto out; + } + + /* We don't have enough room on this page, get a new one. */ + iwl_pcie_free_and_unmap_tso_page(trans, p->page); + +alloc: + p->page = alloc_page(GFP_ATOMIC); + if (!p->page) + return NULL; + p->pos = page_address(p->page); + + info = IWL_TSO_PAGE_INFO(page_address(p->page)); + + /* set the chaining pointer to NULL */ + info->next = NULL; + + /* Create a DMA mapping for the page */ + phys = dma_map_page_attrs(trans->dev, p->page, 0, PAGE_SIZE, + DMA_TO_DEVICE, DMA_ATTR_SKIP_CPU_SYNC); + if (unlikely(dma_mapping_error(trans->dev, phys))) { + __free_page(p->page); + p->page = NULL; + + return NULL; + } + + /* Store physical address and set use count */ + info->dma_addr = phys; + refcount_set(&info->use_count, 1); +out: + *page_ptr = p->page; + /* Return an internal reference for the caller */ + refcount_inc(&info->use_count); + ret = p->pos; + p->pos += len; + + return ret; +} + +/** + * iwl_pcie_get_sgt_tb_phys - Find TB address in mapped SG list + * @sgt: scatter gather table + * @offset: Offset into the mapped memory (i.e. SKB payload data) + * @len: Length of the area + * + * Find the DMA address that corresponds to the SKB payload data at the + * position given by @offset. + * + * Returns: Address for TB entry + */ +dma_addr_t iwl_pcie_get_sgt_tb_phys(struct sg_table *sgt, unsigned int offset, + unsigned int len) +{ + struct scatterlist *sg; + unsigned int sg_offset = 0; + int i; + + /* + * Search the mapped DMA areas in the SG for the area that contains the + * data at offset with the given length. + */ + for_each_sgtable_dma_sg(sgt, sg, i) { + if (offset >= sg_offset && + offset + len <= sg_offset + sg_dma_len(sg)) + return sg_dma_address(sg) + offset - sg_offset; + + sg_offset += sg_dma_len(sg); + } + + WARN_ON_ONCE(1); + + return DMA_MAPPING_ERROR; +} + +/** + * iwl_pcie_prep_tso - Prepare TSO page and SKB for sending + * @trans: transport private data + * @skb: the SKB to map + * @cmd_meta: command meta to store the scatter list information for unmapping + * @hdr: output argument for TSO headers + * @hdr_room: requested length for TSO headers + * + * Allocate space for a scatter gather list and TSO headers and map the SKB + * using the scatter gather list. The SKB is unmapped again when the page is + * free'ed again at the end of the operation. + * + * Returns: newly allocated and mapped scatter gather table with list + */ +struct sg_table *iwl_pcie_prep_tso(struct iwl_trans *trans, struct sk_buff *skb, + struct iwl_cmd_meta *cmd_meta, + u8 **hdr, unsigned int hdr_room) +{ + struct sg_table *sgt; + + if (WARN_ON_ONCE(skb_has_frag_list(skb))) + return NULL; + + *hdr = iwl_pcie_get_page_hdr(trans, + hdr_room + __alignof__(struct sg_table) + + sizeof(struct sg_table) + + (skb_shinfo(skb)->nr_frags + 1) * + sizeof(struct scatterlist), + skb); + if (!*hdr) + return NULL; + + sgt = (void *)PTR_ALIGN(*hdr + hdr_room, __alignof__(struct sg_table)); + sgt->sgl = (void *)(sgt + 1); + + sg_init_table(sgt->sgl, skb_shinfo(skb)->nr_frags + 1); + + /* Only map the data, not the header (it is copied to the TSO page) */ + sgt->orig_nents = skb_to_sgvec(skb, sgt->sgl, skb_headlen(skb), + skb->data_len); + if (WARN_ON_ONCE(sgt->orig_nents <= 0)) + return NULL; + + /* And map the entire SKB */ + if (dma_map_sgtable(trans->dev, sgt, DMA_TO_DEVICE, 0) < 0) + return NULL; + + /* Store non-zero (i.e. valid) offset for unmapping */ + cmd_meta->sg_offset = (unsigned long) sgt & ~PAGE_MASK; + + return sgt; +} + static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb, struct iwl_txq *txq, u8 hdr_len, struct iwl_cmd_meta *out_meta, struct iwl_device_tx_cmd *dev_cmd, u16 tb1_len) { + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload; struct ieee80211_hdr *hdr = (void *)skb->data; unsigned int snap_ip_tcp_hdrlen, ip_hdrlen, total_len, hdr_room; unsigned int mss = skb_shinfo(skb)->gso_size; + unsigned int data_offset = 0; u16 length, iv_len, amsdu_pad; - u8 *start_hdr; - struct iwl_tso_hdr_page *hdr_page; + dma_addr_t start_hdr_phys; + u8 *start_hdr, *pos_hdr; + struct sg_table *sgt; struct tso_t tso; /* if the packet is protected, then it must be CCMP or GCMP */ @@ -1328,7 +1924,7 @@ static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb, trace_iwlwifi_dev_tx(trans->dev, skb, iwl_txq_get_tfd(trans, txq, txq->write_ptr), - trans->txqs.tfd.size, + trans_pcie->txqs.tfd.size, &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len, 0); ip_hdrlen = skb_network_header_len(skb); @@ -1341,13 +1937,14 @@ static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb, (3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr)) + iv_len; /* Our device supports 9 segments at most, it will fit in 1 page */ - hdr_page = get_page_hdr(trans, hdr_room, skb); - if (!hdr_page) + sgt = iwl_pcie_prep_tso(trans, skb, out_meta, &start_hdr, hdr_room); + if (!sgt) return -ENOMEM; - start_hdr = hdr_page->pos; - memcpy(hdr_page->pos, skb->data + hdr_len, iv_len); - hdr_page->pos += iv_len; + start_hdr_phys = iwl_pcie_get_tso_page_phys(start_hdr); + pos_hdr = start_hdr; + memcpy(pos_hdr, skb->data + hdr_len, iv_len); + pos_hdr += iv_len; /* * Pull the ieee80211 header + IV to be able to use TSO core, @@ -1370,45 +1967,43 @@ static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb, min_t(unsigned int, mss, total_len); unsigned int hdr_tb_len; dma_addr_t hdr_tb_phys; - u8 *subf_hdrs_start = hdr_page->pos; + u8 *subf_hdrs_start = pos_hdr; total_len -= data_left; - memset(hdr_page->pos, 0, amsdu_pad); - hdr_page->pos += amsdu_pad; + memset(pos_hdr, 0, amsdu_pad); + pos_hdr += amsdu_pad; amsdu_pad = (4 - (sizeof(struct ethhdr) + snap_ip_tcp_hdrlen + data_left)) & 0x3; - ether_addr_copy(hdr_page->pos, ieee80211_get_DA(hdr)); - hdr_page->pos += ETH_ALEN; - ether_addr_copy(hdr_page->pos, ieee80211_get_SA(hdr)); - hdr_page->pos += ETH_ALEN; + ether_addr_copy(pos_hdr, ieee80211_get_DA(hdr)); + pos_hdr += ETH_ALEN; + ether_addr_copy(pos_hdr, ieee80211_get_SA(hdr)); + pos_hdr += ETH_ALEN; length = snap_ip_tcp_hdrlen + data_left; - *((__be16 *)hdr_page->pos) = cpu_to_be16(length); - hdr_page->pos += sizeof(length); + *((__be16 *)pos_hdr) = cpu_to_be16(length); + pos_hdr += sizeof(length); /* * This will copy the SNAP as well which will be considered * as MAC header. */ - tso_build_hdr(skb, hdr_page->pos, &tso, data_left, !total_len); + tso_build_hdr(skb, pos_hdr, &tso, data_left, !total_len); - hdr_page->pos += snap_ip_tcp_hdrlen; + pos_hdr += snap_ip_tcp_hdrlen; + + hdr_tb_len = pos_hdr - start_hdr; + hdr_tb_phys = iwl_pcie_get_tso_page_phys(start_hdr); - hdr_tb_len = hdr_page->pos - start_hdr; - hdr_tb_phys = dma_map_single(trans->dev, start_hdr, - hdr_tb_len, DMA_TO_DEVICE); - if (unlikely(dma_mapping_error(trans->dev, hdr_tb_phys))) - return -EINVAL; iwl_pcie_txq_build_tfd(trans, txq, hdr_tb_phys, hdr_tb_len, false); trace_iwlwifi_dev_tx_tb(trans->dev, skb, start_hdr, hdr_tb_phys, hdr_tb_len); /* add this subframe's headers' length to the tx_cmd */ - le16_add_cpu(&tx_cmd->len, hdr_page->pos - subf_hdrs_start); + le16_add_cpu(&tx_cmd->len, pos_hdr - subf_hdrs_start); /* prepare the start_hdr for the next subframe */ - start_hdr = hdr_page->pos; + start_hdr = pos_hdr; /* put the payload */ while (data_left) { @@ -1416,9 +2011,9 @@ static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb, data_left); dma_addr_t tb_phys; - tb_phys = dma_map_single(trans->dev, tso.data, - size, DMA_TO_DEVICE); - if (unlikely(dma_mapping_error(trans->dev, tb_phys))) + tb_phys = iwl_pcie_get_sgt_tb_phys(sgt, data_offset, size); + /* Not a real mapping error, use direct comparison */ + if (unlikely(tb_phys == DMA_MAPPING_ERROR)) return -EINVAL; iwl_pcie_txq_build_tfd(trans, txq, tb_phys, @@ -1427,10 +2022,14 @@ static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb, tb_phys, size); data_left -= size; + data_offset += size; tso_build_data(skb, &tso, size); } } + dma_sync_single_for_device(trans->dev, start_hdr_phys, hdr_room, + DMA_TO_DEVICE); + /* re -add the WiFi header and IV */ skb_push(skb, hdr_len + iv_len); @@ -1450,9 +2049,61 @@ static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb, } #endif /* CONFIG_INET */ +#define IWL_TX_CRC_SIZE 4 +#define IWL_TX_DELIMITER_SIZE 4 + +/* + * iwl_txq_gen1_update_byte_cnt_tbl - Set up entry in Tx byte-count array + */ +static void iwl_txq_gen1_update_byte_cnt_tbl(struct iwl_trans *trans, + struct iwl_txq *txq, u16 byte_cnt, + int num_tbs) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwlagn_scd_bc_tbl *scd_bc_tbl; + int write_ptr = txq->write_ptr; + int txq_id = txq->id; + u8 sec_ctl = 0; + u16 len = byte_cnt + IWL_TX_CRC_SIZE + IWL_TX_DELIMITER_SIZE; + __le16 bc_ent; + struct iwl_device_tx_cmd *dev_cmd = txq->entries[txq->write_ptr].cmd; + struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload; + u8 sta_id = tx_cmd->sta_id; + + scd_bc_tbl = trans_pcie->txqs.scd_bc_tbls.addr; + + sec_ctl = tx_cmd->sec_ctl; + + switch (sec_ctl & TX_CMD_SEC_MSK) { + case TX_CMD_SEC_CCM: + len += IEEE80211_CCMP_MIC_LEN; + break; + case TX_CMD_SEC_TKIP: + len += IEEE80211_TKIP_ICV_LEN; + break; + case TX_CMD_SEC_WEP: + len += IEEE80211_WEP_IV_LEN + IEEE80211_WEP_ICV_LEN; + break; + } + if (trans_pcie->txqs.bc_table_dword) + len = DIV_ROUND_UP(len, 4); + + if (WARN_ON(len > 0xFFF || write_ptr >= TFD_QUEUE_SIZE_MAX)) + return; + + bc_ent = cpu_to_le16(len | (sta_id << 12)); + + scd_bc_tbl[txq_id].tfd_offset[write_ptr] = bc_ent; + + if (write_ptr < TFD_QUEUE_SIZE_BC_DUP) + scd_bc_tbl[txq_id].tfd_offset[TFD_QUEUE_SIZE_MAX + write_ptr] = + bc_ent; +} + int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb, struct iwl_device_tx_cmd *dev_cmd, int txq_id) { + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); struct ieee80211_hdr *hdr; struct iwl_tx_cmd *tx_cmd = (struct iwl_tx_cmd *)dev_cmd->payload; struct iwl_cmd_meta *out_meta; @@ -1467,14 +2118,14 @@ int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb, u16 wifi_seq; bool amsdu; - txq = trans->txqs.txq[txq_id]; + txq = trans_pcie->txqs.txq[txq_id]; - if (WARN_ONCE(!test_bit(txq_id, trans->txqs.queue_used), + if (WARN_ONCE(!test_bit(txq_id, trans_pcie->txqs.queue_used), "TX on unused queue %d\n", txq_id)) return -EINVAL; if (skb_is_nonlinear(skb) && - skb_shinfo(skb)->nr_frags > IWL_TRANS_MAX_FRAGS(trans) && + skb_shinfo(skb)->nr_frags > IWL_TRANS_PCIE_MAX_FRAGS(trans_pcie) && __skb_linearize(skb)) return -ENOMEM; @@ -1495,7 +2146,7 @@ int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb, struct iwl_device_tx_cmd **dev_cmd_ptr; dev_cmd_ptr = (void *)((u8 *)skb->cb + - trans->txqs.dev_cmd_offs); + trans_pcie->txqs.dev_cmd_offs); *dev_cmd_ptr = dev_cmd; __skb_queue_tail(&txq->overflow_q, skb); @@ -1533,7 +2184,7 @@ int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb, /* Set up first empty entry in queue's array of Tx/cmd buffers */ out_meta = &txq->entries[txq->write_ptr].meta; - out_meta->flags = 0; + memset(out_meta, 0, sizeof(*out_meta)); /* * The second TB (tb1) points to the remainder of the TX command @@ -1578,7 +2229,7 @@ int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb, trace_iwlwifi_dev_tx(trans->dev, skb, iwl_txq_get_tfd(trans, txq, txq->write_ptr), - trans->txqs.tfd.size, + trans_pcie->txqs.tfd.size, &dev_cmd->hdr, IWL_FIRST_TB_SIZE + tb1_len, hdr_len); @@ -1613,8 +2264,7 @@ int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb, tfd = iwl_txq_get_tfd(trans, txq, txq->write_ptr); /* Set up entry for this TFD in Tx byte-count array */ iwl_txq_gen1_update_byte_cnt_tbl(trans, txq, le16_to_cpu(tx_cmd->len), - iwl_txq_gen1_tfd_get_num_tbs(trans, - tfd)); + iwl_txq_gen1_tfd_get_num_tbs(tfd)); wait_write_ptr = ieee80211_has_morefrags(fc); @@ -1649,3 +2299,379 @@ out_err: spin_unlock(&txq->lock); return -1; } + +static void iwl_txq_gen1_inval_byte_cnt_tbl(struct iwl_trans *trans, + struct iwl_txq *txq, + int read_ptr) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwlagn_scd_bc_tbl *scd_bc_tbl = trans_pcie->txqs.scd_bc_tbls.addr; + int txq_id = txq->id; + u8 sta_id = 0; + __le16 bc_ent; + struct iwl_device_tx_cmd *dev_cmd = txq->entries[read_ptr].cmd; + struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload; + + WARN_ON(read_ptr >= TFD_QUEUE_SIZE_MAX); + + if (txq_id != trans_pcie->txqs.cmd.q_id) + sta_id = tx_cmd->sta_id; + + bc_ent = cpu_to_le16(1 | (sta_id << 12)); + + scd_bc_tbl[txq_id].tfd_offset[read_ptr] = bc_ent; + + if (read_ptr < TFD_QUEUE_SIZE_BC_DUP) + scd_bc_tbl[txq_id].tfd_offset[TFD_QUEUE_SIZE_MAX + read_ptr] = + bc_ent; +} + +/* Frees buffers until index _not_ inclusive */ +void iwl_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn, + struct sk_buff_head *skbs, bool is_flush) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; + int tfd_num, read_ptr, last_to_free; + int txq_read_ptr, txq_write_ptr; + + /* This function is not meant to release cmd queue*/ + if (WARN_ON(txq_id == trans_pcie->txqs.cmd.q_id)) + return; + + if (WARN_ON(!txq)) + return; + + tfd_num = iwl_txq_get_cmd_index(txq, ssn); + + spin_lock_bh(&txq->reclaim_lock); + + spin_lock(&txq->lock); + txq_read_ptr = txq->read_ptr; + txq_write_ptr = txq->write_ptr; + spin_unlock(&txq->lock); + + read_ptr = iwl_txq_get_cmd_index(txq, txq_read_ptr); + + if (!test_bit(txq_id, trans_pcie->txqs.queue_used)) { + IWL_DEBUG_TX_QUEUES(trans, "Q %d inactive - ignoring idx %d\n", + txq_id, ssn); + goto out; + } + + if (read_ptr == tfd_num) + goto out; + + IWL_DEBUG_TX_REPLY(trans, "[Q %d] %d (%d) -> %d (%d)\n", + txq_id, read_ptr, txq_read_ptr, tfd_num, ssn); + + /* Since we free until index _not_ inclusive, the one before index is + * the last we will free. This one must be used + */ + last_to_free = iwl_txq_dec_wrap(trans, tfd_num); + + if (!iwl_txq_used(txq, last_to_free, txq_read_ptr, txq_write_ptr)) { + IWL_ERR(trans, + "%s: Read index for txq id (%d), last_to_free %d is out of range [0-%d] %d %d.\n", + __func__, txq_id, last_to_free, + trans->trans_cfg->base_params->max_tfd_queue_size, + txq_write_ptr, txq_read_ptr); + + iwl_op_mode_time_point(trans->op_mode, + IWL_FW_INI_TIME_POINT_FAKE_TX, + NULL); + goto out; + } + + if (WARN_ON(!skb_queue_empty(skbs))) + goto out; + + for (; + read_ptr != tfd_num; + txq_read_ptr = iwl_txq_inc_wrap(trans, txq_read_ptr), + read_ptr = iwl_txq_get_cmd_index(txq, txq_read_ptr)) { + struct iwl_cmd_meta *cmd_meta = &txq->entries[read_ptr].meta; + struct sk_buff *skb = txq->entries[read_ptr].skb; + + if (WARN_ONCE(!skb, "no SKB at %d (%d) on queue %d\n", + read_ptr, txq_read_ptr, txq_id)) + continue; + + iwl_pcie_free_tso_pages(trans, skb, cmd_meta); + + __skb_queue_tail(skbs, skb); + + txq->entries[read_ptr].skb = NULL; + + if (!trans->trans_cfg->gen2) + iwl_txq_gen1_inval_byte_cnt_tbl(trans, txq, + txq_read_ptr); + + iwl_txq_free_tfd(trans, txq, txq_read_ptr); + } + + spin_lock(&txq->lock); + txq->read_ptr = txq_read_ptr; + + iwl_txq_progress(txq); + + if (iwl_txq_space(trans, txq) > txq->low_mark && + test_bit(txq_id, trans_pcie->txqs.queue_stopped)) { + struct sk_buff_head overflow_skbs; + struct sk_buff *skb; + + __skb_queue_head_init(&overflow_skbs); + skb_queue_splice_init(&txq->overflow_q, + is_flush ? skbs : &overflow_skbs); + + /* + * We are going to transmit from the overflow queue. + * Remember this state so that wait_for_txq_empty will know we + * are adding more packets to the TFD queue. It cannot rely on + * the state of &txq->overflow_q, as we just emptied it, but + * haven't TXed the content yet. + */ + txq->overflow_tx = true; + + /* + * This is tricky: we are in reclaim path and are holding + * reclaim_lock, so noone will try to access the txq data + * from that path. We stopped tx, so we can't have tx as well. + * Bottom line, we can unlock and re-lock later. + */ + spin_unlock(&txq->lock); + + while ((skb = __skb_dequeue(&overflow_skbs))) { + struct iwl_device_tx_cmd *dev_cmd_ptr; + + dev_cmd_ptr = *(void **)((u8 *)skb->cb + + trans_pcie->txqs.dev_cmd_offs); + + /* + * Note that we can very well be overflowing again. + * In that case, iwl_txq_space will be small again + * and we won't wake mac80211's queue. + */ + iwl_trans_tx(trans, skb, dev_cmd_ptr, txq_id); + } + + if (iwl_txq_space(trans, txq) > txq->low_mark) + iwl_trans_pcie_wake_queue(trans, txq); + + spin_lock(&txq->lock); + txq->overflow_tx = false; + } + + spin_unlock(&txq->lock); +out: + spin_unlock_bh(&txq->reclaim_lock); +} + +/* Set wr_ptr of specific device and txq */ +void iwl_pcie_set_q_ptrs(struct iwl_trans *trans, int txq_id, int ptr) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + struct iwl_txq *txq = trans_pcie->txqs.txq[txq_id]; + + spin_lock_bh(&txq->lock); + + txq->write_ptr = ptr; + txq->read_ptr = txq->write_ptr; + + spin_unlock_bh(&txq->lock); +} + +void iwl_pcie_freeze_txq_timer(struct iwl_trans *trans, + unsigned long txqs, bool freeze) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + int queue; + + for_each_set_bit(queue, &txqs, BITS_PER_LONG) { + struct iwl_txq *txq = trans_pcie->txqs.txq[queue]; + unsigned long now; + + spin_lock_bh(&txq->lock); + + now = jiffies; + + if (txq->frozen == freeze) + goto next_queue; + + IWL_DEBUG_TX_QUEUES(trans, "%s TXQ %d\n", + freeze ? "Freezing" : "Waking", queue); + + txq->frozen = freeze; + + if (txq->read_ptr == txq->write_ptr) + goto next_queue; + + if (freeze) { + if (unlikely(time_after(now, + txq->stuck_timer.expires))) { + /* + * The timer should have fired, maybe it is + * spinning right now on the lock. + */ + goto next_queue; + } + /* remember how long until the timer fires */ + txq->frozen_expiry_remainder = + txq->stuck_timer.expires - now; + del_timer(&txq->stuck_timer); + goto next_queue; + } + + /* + * Wake a non-empty queue -> arm timer with the + * remainder before it froze + */ + mod_timer(&txq->stuck_timer, + now + txq->frozen_expiry_remainder); + +next_queue: + spin_unlock_bh(&txq->lock); + } +} + +#define HOST_COMPLETE_TIMEOUT (2 * HZ) + +static int iwl_trans_pcie_send_hcmd_sync(struct iwl_trans *trans, + struct iwl_host_cmd *cmd) +{ + struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans); + const char *cmd_str = iwl_get_cmd_string(trans, cmd->id); + struct iwl_txq *txq = trans_pcie->txqs.txq[trans_pcie->txqs.cmd.q_id]; + int cmd_idx; + int ret; + + IWL_DEBUG_INFO(trans, "Attempting to send sync command %s\n", cmd_str); + + if (WARN(test_and_set_bit(STATUS_SYNC_HCMD_ACTIVE, + &trans->status), + "Command %s: a command is already active!\n", cmd_str)) + return -EIO; + + IWL_DEBUG_INFO(trans, "Setting HCMD_ACTIVE for command %s\n", cmd_str); + + if (trans->trans_cfg->gen2) + cmd_idx = iwl_pcie_gen2_enqueue_hcmd(trans, cmd); + else + cmd_idx = iwl_pcie_enqueue_hcmd(trans, cmd); + + if (cmd_idx < 0) { + ret = cmd_idx; + clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status); + IWL_ERR(trans, "Error sending %s: enqueue_hcmd failed: %d\n", + cmd_str, ret); + return ret; + } + + ret = wait_event_timeout(trans->wait_command_queue, + !test_bit(STATUS_SYNC_HCMD_ACTIVE, + &trans->status), + HOST_COMPLETE_TIMEOUT); + if (!ret) { + IWL_ERR(trans, "Error sending %s: time out after %dms.\n", + cmd_str, jiffies_to_msecs(HOST_COMPLETE_TIMEOUT)); + + IWL_ERR(trans, "Current CMD queue read_ptr %d write_ptr %d\n", + txq->read_ptr, txq->write_ptr); + + clear_bit(STATUS_SYNC_HCMD_ACTIVE, &trans->status); + IWL_DEBUG_INFO(trans, "Clearing HCMD_ACTIVE for command %s\n", + cmd_str); + ret = -ETIMEDOUT; + + iwl_trans_sync_nmi(trans); + goto cancel; + } + + if (test_bit(STATUS_FW_ERROR, &trans->status)) { + if (!test_and_clear_bit(STATUS_SUPPRESS_CMD_ERROR_ONCE, + &trans->status)) { + IWL_ERR(trans, "FW error in SYNC CMD %s\n", cmd_str); + dump_stack(); + } + ret = -EIO; + goto cancel; + } + + if (!(cmd->flags & CMD_SEND_IN_RFKILL) && + test_bit(STATUS_RFKILL_OPMODE, &trans->status)) { + IWL_DEBUG_RF_KILL(trans, "RFKILL in SYNC CMD... no rsp\n"); + ret = -ERFKILL; + goto cancel; + } + + if ((cmd->flags & CMD_WANT_SKB) && !cmd->resp_pkt) { + IWL_ERR(trans, "Error: Response NULL in '%s'\n", cmd_str); + ret = -EIO; + goto cancel; + } + + return 0; + +cancel: + if (cmd->flags & CMD_WANT_SKB) { + /* + * Cancel the CMD_WANT_SKB flag for the cmd in the + * TX cmd queue. Otherwise in case the cmd comes + * in later, it will possibly set an invalid + * address (cmd->meta.source). + */ + txq->entries[cmd_idx].meta.flags &= ~CMD_WANT_SKB; + } + + if (cmd->resp_pkt) { + iwl_free_resp(cmd); + cmd->resp_pkt = NULL; + } + + return ret; +} + +int iwl_trans_pcie_send_hcmd(struct iwl_trans *trans, + struct iwl_host_cmd *cmd) +{ + /* Make sure the NIC is still alive in the bus */ + if (test_bit(STATUS_TRANS_DEAD, &trans->status)) + return -ENODEV; + + if (!(cmd->flags & CMD_SEND_IN_RFKILL) && + test_bit(STATUS_RFKILL_OPMODE, &trans->status)) { + IWL_DEBUG_RF_KILL(trans, "Dropping CMD 0x%x: RF KILL\n", + cmd->id); + return -ERFKILL; + } + + if (unlikely(trans->system_pm_mode == IWL_PLAT_PM_MODE_D3 && + !(cmd->flags & CMD_SEND_IN_D3))) { + IWL_DEBUG_WOWLAN(trans, "Dropping CMD 0x%x: D3\n", cmd->id); + return -EHOSTDOWN; + } + + if (cmd->flags & CMD_ASYNC) { + int ret; + + /* An asynchronous command can not expect an SKB to be set. */ + if (WARN_ON(cmd->flags & CMD_WANT_SKB)) + return -EINVAL; + + if (trans->trans_cfg->gen2) + ret = iwl_pcie_gen2_enqueue_hcmd(trans, cmd); + else + ret = iwl_pcie_enqueue_hcmd(trans, cmd); + + if (ret < 0) { + IWL_ERR(trans, + "Error sending %s: enqueue_hcmd failed: %d\n", + iwl_get_cmd_string(trans, cmd->id), ret); + return ret; + } + return 0; + } + + return iwl_trans_pcie_send_hcmd_sync(trans, cmd); +} +IWL_EXPORT_SYMBOL(iwl_trans_pcie_send_hcmd); |