Merge tag 'iwlwifi-for-kalle-2020-01-11' of...

Merge tag 'iwlwifi-for-kalle-2020-01-11' of git://git.kernel.org/pub/scm/linux/kernel/git/iwlwifi/iwlwifi-fixes

First batch of fixes intended for v5.5

* Don't send the PPAG command when PPAG is disabled, since it can
  cause problems;
* A few fixes for a HW bug;
* A fix for RS offload;
* A fix for 3168 devices where the NVM tables where the wrong tables
  were being read.
* Fix a couple of potential memory leaks in TXQ code;
* Disable L0S states in all hardware since our hardware doesn't
  officially support them anymore (and older versions of the hardware
  had instability in these states);
* Remove lar_disable parameter since it has been causing issues for
  some people who erroneously disable it;
* Force the debug monitor HW to stop also when debug is disabled,
  since it sometimes stays on and prevents low system power states;

drivers/net/wireless/intel/iwlwifi/dvm/tx.c

@@ -267,7 +267,7 @@ int iwlagn_tx_skb(struct iwl_priv *priv,
 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
 	struct iwl_station_priv *sta_priv = NULL;
 	struct iwl_rxon_context *ctx = &priv->contexts[IWL_RXON_CTX_BSS];
-	struct iwl_device_cmd *dev_cmd;
+	struct iwl_device_tx_cmd *dev_cmd;
 	struct iwl_tx_cmd *tx_cmd;
 	__le16 fc;
 	u8 hdr_len;
@@ -348,7 +348,6 @@ int iwlagn_tx_skb(struct iwl_priv *priv,
 	if (unlikely(!dev_cmd))
 		goto drop_unlock_priv;
 
-	memset(dev_cmd, 0, sizeof(*dev_cmd));
 	dev_cmd->hdr.cmd = REPLY_TX;
 	tx_cmd = (struct iwl_tx_cmd *) dev_cmd->payload;
 

drivers/net/wireless/intel/iwlwifi/fw/acpi.c

@@ -357,8 +357,8 @@ int iwl_sar_get_ewrd_table(struct iwl_fw_runtime *fwrt)
 {
 	union acpi_object *wifi_pkg, *data;
 	bool enabled;
-	int i, n_profiles, tbl_rev;
-	int  ret = 0;
+	int i, n_profiles, tbl_rev, pos;
+	int ret = 0;
 
 	data = iwl_acpi_get_object(fwrt->dev, ACPI_EWRD_METHOD);
 	if (IS_ERR(data))
@@ -390,10 +390,10 @@ int iwl_sar_get_ewrd_table(struct iwl_fw_runtime *fwrt)
 		goto out_free;
 	}
 
-	for (i = 0; i < n_profiles; i++) {
-		/* the tables start at element 3 */
-		int pos = 3;
+	/* the tables start at element 3 */
+	pos = 3;
 
+	for (i = 0; i < n_profiles; i++) {
 		/* The EWRD profiles officially go from 2 to 4, but we
 		 * save them in sar_profiles[1-3] (because we don't
 		 * have profile 0).  So in the array we start from 1.

drivers/net/wireless/intel/iwlwifi/fw/dbg.c

@@ -2669,12 +2669,7 @@ int iwl_fw_dbg_stop_restart_recording(struct iwl_fw_runtime *fwrt,
 {
 	int ret = 0;
 
-	/* if the FW crashed or not debug monitor cfg was given, there is
-	 * no point in changing the recording state
-	 */
-	if (test_bit(STATUS_FW_ERROR, &fwrt->trans->status) ||
-	    (!fwrt->trans->dbg.dest_tlv &&
-	     fwrt->trans->dbg.ini_dest == IWL_FW_INI_LOCATION_INVALID))
+	if (test_bit(STATUS_FW_ERROR, &fwrt->trans->status))
 		return 0;
 
 	if (fw_has_capa(&fwrt->fw->ucode_capa,

drivers/net/wireless/intel/iwlwifi/iwl-csr.h

@@ -379,7 +379,7 @@ enum {
 
 
 /* CSR GIO */
-#define CSR_GIO_REG_VAL_L0S_ENABLED	(0x00000002)
+#define CSR_GIO_REG_VAL_L0S_DISABLED	(0x00000002)
 
 /*
  * UCODE-DRIVER GP (general purpose) mailbox register 1

drivers/net/wireless/intel/iwlwifi/iwl-dbg-tlv.c

@@ -480,7 +480,14 @@ static int iwl_dbg_tlv_alloc_fragment(struct iwl_fw_runtime *fwrt,
 	if (!frag || frag->size || !pages)
 		return -EIO;
 
-	while (pages) {
+	/*
+	 * We try to allocate as many pages as we can, starting with
+	 * the requested amount and going down until we can allocate
+	 * something.  Because of DIV_ROUND_UP(), pages will never go
+	 * down to 0 and stop the loop, so stop when pages reaches 1,
+	 * which is too small anyway.
+	 */
+	while (pages > 1) {
 		block = dma_alloc_coherent(fwrt->dev, pages * PAGE_SIZE,
 					   &physical,
 					   GFP_KERNEL | __GFP_NOWARN);

drivers/net/wireless/intel/iwlwifi/iwl-drv.c

@@ -1817,9 +1817,6 @@ MODULE_PARM_DESC(antenna_coupling,
 module_param_named(nvm_file, iwlwifi_mod_params.nvm_file, charp, 0444);
 MODULE_PARM_DESC(nvm_file, "NVM file name");
 
-module_param_named(lar_disable, iwlwifi_mod_params.lar_disable, bool, 0444);
-MODULE_PARM_DESC(lar_disable, "disable LAR functionality (default: N)");
-
 module_param_named(uapsd_disable, iwlwifi_mod_params.uapsd_disable, uint, 0644);
 MODULE_PARM_DESC(uapsd_disable,
 		 "disable U-APSD functionality bitmap 1: BSS 2: P2P Client (default: 3)");

drivers/net/wireless/intel/iwlwifi/iwl-modparams.h

@@ -115,7 +115,6 @@ enum iwl_uapsd_disable {
  * @nvm_file: specifies a external NVM file
  * @uapsd_disable: disable U-APSD, see &enum iwl_uapsd_disable, default =
  *	IWL_DISABLE_UAPSD_BSS | IWL_DISABLE_UAPSD_P2P_CLIENT
- * @lar_disable: disable LAR (regulatory), default = 0
  * @fw_monitor: allow to use firmware monitor
  * @disable_11ac: disable VHT capabilities, default = false.
  * @remove_when_gone: remove an inaccessible device from the PCIe bus.
@@ -136,7 +135,6 @@ struct iwl_mod_params {
 	int antenna_coupling;
 	char *nvm_file;
 	u32 uapsd_disable;
-	bool lar_disable;
 	bool fw_monitor;
 	bool disable_11ac;
 	/**

drivers/net/wireless/intel/iwlwifi/iwl-nvm-parse.c

@@ -224,6 +224,34 @@ enum iwl_nvm_channel_flags {
 	NVM_CHANNEL_DC_HIGH		= BIT(12),
 };
 
+/**
+ * enum iwl_reg_capa_flags - global flags applied for the whole regulatory
+ * domain.
+ * @REG_CAPA_BF_CCD_LOW_BAND: Beam-forming or Cyclic Delay Diversity in the
+ *	2.4Ghz band is allowed.
+ * @REG_CAPA_BF_CCD_HIGH_BAND: Beam-forming or Cyclic Delay Diversity in the
+ *	5Ghz band is allowed.
+ * @REG_CAPA_160MHZ_ALLOWED: 11ac channel with a width of 160Mhz is allowed
+ *	for this regulatory domain (valid only in 5Ghz).
+ * @REG_CAPA_80MHZ_ALLOWED: 11ac channel with a width of 80Mhz is allowed
+ *	for this regulatory domain (valid only in 5Ghz).
+ * @REG_CAPA_MCS_8_ALLOWED: 11ac with MCS 8 is allowed.
+ * @REG_CAPA_MCS_9_ALLOWED: 11ac with MCS 9 is allowed.
+ * @REG_CAPA_40MHZ_FORBIDDEN: 11n channel with a width of 40Mhz is forbidden
+ *	for this regulatory domain (valid only in 5Ghz).
+ * @REG_CAPA_DC_HIGH_ENABLED: DC HIGH allowed.
+ */
+enum iwl_reg_capa_flags {
+	REG_CAPA_BF_CCD_LOW_BAND	= BIT(0),
+	REG_CAPA_BF_CCD_HIGH_BAND	= BIT(1),
+	REG_CAPA_160MHZ_ALLOWED		= BIT(2),
+	REG_CAPA_80MHZ_ALLOWED		= BIT(3),
+	REG_CAPA_MCS_8_ALLOWED		= BIT(4),
+	REG_CAPA_MCS_9_ALLOWED		= BIT(5),
+	REG_CAPA_40MHZ_FORBIDDEN	= BIT(7),
+	REG_CAPA_DC_HIGH_ENABLED	= BIT(9),
+};
+
 static inline void iwl_nvm_print_channel_flags(struct device *dev, u32 level,
 					       int chan, u32 flags)
 {
@@ -939,10 +967,11 @@ iwl_nvm_no_wide_in_5ghz(struct iwl_trans *trans, const struct iwl_cfg *cfg,
 
 struct iwl_nvm_data *
 iwl_parse_nvm_data(struct iwl_trans *trans, const struct iwl_cfg *cfg,
+		   const struct iwl_fw *fw,
 		   const __be16 *nvm_hw, const __le16 *nvm_sw,
 		   const __le16 *nvm_calib, const __le16 *regulatory,
 		   const __le16 *mac_override, const __le16 *phy_sku,
-		   u8 tx_chains, u8 rx_chains, bool lar_fw_supported)
+		   u8 tx_chains, u8 rx_chains)
 {
 	struct iwl_nvm_data *data;
 	bool lar_enabled;
@@ -1022,7 +1051,8 @@ iwl_parse_nvm_data(struct iwl_trans *trans, const struct iwl_cfg *cfg,
 		return NULL;
 	}
 
-	if (lar_fw_supported && lar_enabled)
+	if (lar_enabled &&
+	    fw_has_capa(&fw->ucode_capa, IWL_UCODE_TLV_CAPA_LAR_SUPPORT))
 		sbands_flags |= IWL_NVM_SBANDS_FLAGS_LAR;
 
 	if (iwl_nvm_no_wide_in_5ghz(trans, cfg, nvm_hw))
@@ -1038,6 +1068,7 @@ IWL_EXPORT_SYMBOL(iwl_parse_nvm_data);
 
 static u32 iwl_nvm_get_regdom_bw_flags(const u16 *nvm_chan,
 				       int ch_idx, u16 nvm_flags,
+				       u16 cap_flags,
 				       const struct iwl_cfg *cfg)
 {
 	u32 flags = NL80211_RRF_NO_HT40;
@@ -1076,13 +1107,27 @@ static u32 iwl_nvm_get_regdom_bw_flags(const u16 *nvm_chan,
 	    (flags & NL80211_RRF_NO_IR))
 		flags |= NL80211_RRF_GO_CONCURRENT;
 
+	/*
+	 * cap_flags is per regulatory domain so apply it for every channel
+	 */
+	if (ch_idx >= NUM_2GHZ_CHANNELS) {
+		if (cap_flags & REG_CAPA_40MHZ_FORBIDDEN)
+			flags |= NL80211_RRF_NO_HT40;
+
+		if (!(cap_flags & REG_CAPA_80MHZ_ALLOWED))
+			flags |= NL80211_RRF_NO_80MHZ;
+
+		if (!(cap_flags & REG_CAPA_160MHZ_ALLOWED))
+			flags |= NL80211_RRF_NO_160MHZ;
+	}
+
 	return flags;
 }
 
 struct ieee80211_regdomain *
 iwl_parse_nvm_mcc_info(struct device *dev, const struct iwl_cfg *cfg,
 		       int num_of_ch, __le32 *channels, u16 fw_mcc,
-		       u16 geo_info)
+		       u16 geo_info, u16 cap)
 {
 	int ch_idx;
 	u16 ch_flags;
@@ -1140,7 +1185,8 @@ iwl_parse_nvm_mcc_info(struct device *dev, const struct iwl_cfg *cfg,
 		}
 
 		reg_rule_flags = iwl_nvm_get_regdom_bw_flags(nvm_chan, ch_idx,
-							     ch_flags, cfg);
+							     ch_flags, cap,
+							     cfg);
 
 		/* we can't continue the same rule */
 		if (ch_idx == 0 || prev_reg_rule_flags != reg_rule_flags ||
@@ -1405,9 +1451,6 @@ struct iwl_nvm_data *iwl_get_nvm(struct iwl_trans *trans,
 		.id = WIDE_ID(REGULATORY_AND_NVM_GROUP, NVM_GET_INFO)
 	};
 	int  ret;
-	bool lar_fw_supported = !iwlwifi_mod_params.lar_disable &&
-				fw_has_capa(&fw->ucode_capa,
-					    IWL_UCODE_TLV_CAPA_LAR_SUPPORT);
 	bool empty_otp;
 	u32 mac_flags;
 	u32 sbands_flags = 0;
@@ -1485,7 +1528,9 @@ struct iwl_nvm_data *iwl_get_nvm(struct iwl_trans *trans,
 	nvm->valid_tx_ant = (u8)le32_to_cpu(rsp->phy_sku.tx_chains);
 	nvm->valid_rx_ant = (u8)le32_to_cpu(rsp->phy_sku.rx_chains);
 
-	if (le32_to_cpu(rsp->regulatory.lar_enabled) && lar_fw_supported) {
+	if (le32_to_cpu(rsp->regulatory.lar_enabled) &&
+	    fw_has_capa(&fw->ucode_capa,
+			IWL_UCODE_TLV_CAPA_LAR_SUPPORT)) {
 		nvm->lar_enabled = true;
 		sbands_flags |= IWL_NVM_SBANDS_FLAGS_LAR;
 	}

drivers/net/wireless/intel/iwlwifi/iwl-nvm-parse.h

@@ -7,7 +7,7 @@
  *
  * Copyright(c) 2008 - 2015 Intel Corporation. All rights reserved.
  * Copyright(c) 2016 - 2017 Intel Deutschland GmbH
- * Copyright(c) 2018        Intel Corporation
+ * Copyright(c) 2018 - 2019 Intel Corporation
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of version 2 of the GNU General Public License as
@@ -29,7 +29,7 @@
  *
  * Copyright(c) 2005 - 2014 Intel Corporation. All rights reserved.
  * Copyright(c) 2016 - 2017 Intel Deutschland GmbH
- * Copyright(c) 2018        Intel Corporation
+ * Copyright(c) 2018 - 2019 Intel Corporation
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -85,10 +85,11 @@ enum iwl_nvm_sbands_flags {
  */
 struct iwl_nvm_data *
 iwl_parse_nvm_data(struct iwl_trans *trans, const struct iwl_cfg *cfg,
+		   const struct iwl_fw *fw,
 		   const __be16 *nvm_hw, const __le16 *nvm_sw,
 		   const __le16 *nvm_calib, const __le16 *regulatory,
 		   const __le16 *mac_override, const __le16 *phy_sku,
-		   u8 tx_chains, u8 rx_chains, bool lar_fw_supported);
+		   u8 tx_chains, u8 rx_chains);
 
 /**
  * iwl_parse_mcc_info - parse MCC (mobile country code) info coming from FW
@@ -103,7 +104,7 @@ iwl_parse_nvm_data(struct iwl_trans *trans, const struct iwl_cfg *cfg,
 struct ieee80211_regdomain *
 iwl_parse_nvm_mcc_info(struct device *dev, const struct iwl_cfg *cfg,
 		       int num_of_ch, __le32 *channels, u16 fw_mcc,
-		       u16 geo_info);
+		       u16 geo_info, u16 cap);
 
 /**
  * struct iwl_nvm_section - describes an NVM section in memory.

drivers/net/wireless/intel/iwlwifi/iwl-trans.c

@@ -66,7 +66,9 @@
 
 struct iwl_trans *iwl_trans_alloc(unsigned int priv_size,
 				  struct device *dev,
-				  const struct iwl_trans_ops *ops)
+				  const struct iwl_trans_ops *ops,
+				  unsigned int cmd_pool_size,
+				  unsigned int cmd_pool_align)
 {
 	struct iwl_trans *trans;
 #ifdef CONFIG_LOCKDEP
@@ -90,10 +92,8 @@ struct iwl_trans *iwl_trans_alloc(unsigned int priv_size,
 		 "iwl_cmd_pool:%s", dev_name(trans->dev));
 	trans->dev_cmd_pool =
 		kmem_cache_create(trans->dev_cmd_pool_name,
-				  sizeof(struct iwl_device_cmd),
-				  sizeof(void *),
-				  SLAB_HWCACHE_ALIGN,
-				  NULL);
+				  cmd_pool_size, cmd_pool_align,
+				  SLAB_HWCACHE_ALIGN, NULL);
 	if (!trans->dev_cmd_pool)
 		return NULL;
 

drivers/net/wireless/intel/iwlwifi/iwl-trans.h

@@ -193,6 +193,18 @@ struct iwl_device_cmd {
 	};
 } __packed;
 
+/**
+ * struct iwl_device_tx_cmd - buffer for TX command
+ * @hdr: the header
+ * @payload: the payload placeholder
+ *
+ * The actual structure is sized dynamically according to need.
+ */
+struct iwl_device_tx_cmd {
+	struct iwl_cmd_header hdr;
+	u8 payload[];
+} __packed;
+
 #define TFD_MAX_PAYLOAD_SIZE (sizeof(struct iwl_device_cmd))
 
 /*
@@ -544,7 +556,7 @@ struct iwl_trans_ops {
 	int (*send_cmd)(struct iwl_trans *trans, struct iwl_host_cmd *cmd);
 
 	int (*tx)(struct iwl_trans *trans, struct sk_buff *skb,
-		  struct iwl_device_cmd *dev_cmd, int queue);
+		  struct iwl_device_tx_cmd *dev_cmd, int queue);
 	void (*reclaim)(struct iwl_trans *trans, int queue, int ssn,
 			struct sk_buff_head *skbs);
 
@@ -948,22 +960,22 @@ iwl_trans_dump_data(struct iwl_trans *trans, u32 dump_mask)
 	return trans->ops->dump_data(trans, dump_mask);
 }
 
-static inline struct iwl_device_cmd *
+static inline struct iwl_device_tx_cmd *
 iwl_trans_alloc_tx_cmd(struct iwl_trans *trans)
 {
-	return kmem_cache_alloc(trans->dev_cmd_pool, GFP_ATOMIC);
+	return kmem_cache_zalloc(trans->dev_cmd_pool, GFP_ATOMIC);
 }
 
 int iwl_trans_send_cmd(struct iwl_trans *trans, struct iwl_host_cmd *cmd);
 
 static inline void iwl_trans_free_tx_cmd(struct iwl_trans *trans,
-					 struct iwl_device_cmd *dev_cmd)
+					 struct iwl_device_tx_cmd *dev_cmd)
 {
 	kmem_cache_free(trans->dev_cmd_pool, dev_cmd);
 }
 
 static inline int iwl_trans_tx(struct iwl_trans *trans, struct sk_buff *skb,
-			       struct iwl_device_cmd *dev_cmd, int queue)
+			       struct iwl_device_tx_cmd *dev_cmd, int queue)
 {
 	if (unlikely(test_bit(STATUS_FW_ERROR, &trans->status)))
 		return -EIO;
@@ -1271,7 +1283,9 @@ static inline bool iwl_trans_dbg_ini_valid(struct iwl_trans *trans)
  *****************************************************/
 struct iwl_trans *iwl_trans_alloc(unsigned int priv_size,
 				  struct device *dev,
-				  const struct iwl_trans_ops *ops);
+				  const struct iwl_trans_ops *ops,
+				  unsigned int cmd_pool_size,
+				  unsigned int cmd_pool_align);
 void iwl_trans_free(struct iwl_trans *trans);
 
 /*****************************************************

drivers/net/wireless/intel/iwlwifi/mvm/fw.c

@@ -841,9 +841,13 @@ int iwl_mvm_ppag_send_cmd(struct iwl_mvm *mvm)
 		return 0;
 	}
 
+	if (!mvm->fwrt.ppag_table.enabled) {
+		IWL_DEBUG_RADIO(mvm,
+				"PPAG not enabled, command not sent.\n");
+		return 0;
+	}
+
 	IWL_DEBUG_RADIO(mvm, "Sending PER_PLATFORM_ANT_GAIN_CMD\n");
-	IWL_DEBUG_RADIO(mvm, "PPAG is %s\n",
-			mvm->fwrt.ppag_table.enabled ? "enabled" : "disabled");
 
 	for (i = 0; i < ACPI_PPAG_NUM_CHAINS; i++) {
 		for (j = 0; j < ACPI_PPAG_NUM_SUB_BANDS; j++) {

drivers/net/wireless/intel/iwlwifi/mvm/mac80211.c

@@ -256,7 +256,8 @@ struct ieee80211_regdomain *iwl_mvm_get_regdomain(struct wiphy *wiphy,
 				      __le32_to_cpu(resp->n_channels),
 				      resp->channels,
 				      __le16_to_cpu(resp->mcc),
-				      __le16_to_cpu(resp->geo_info));
+				      __le16_to_cpu(resp->geo_info),
+				      __le16_to_cpu(resp->cap));
 	/* Store the return source id */
 	src_id = resp->source_id;
 	kfree(resp);
@@ -754,6 +755,20 @@ int iwl_mvm_mac_setup_register(struct iwl_mvm *mvm)
 	return ret;
 }
 
+static void iwl_mvm_tx_skb(struct iwl_mvm *mvm, struct sk_buff *skb,
+			   struct ieee80211_sta *sta)
+{
+	if (likely(sta)) {
+		if (likely(iwl_mvm_tx_skb_sta(mvm, skb, sta) == 0))
+			return;
+	} else {
+		if (likely(iwl_mvm_tx_skb_non_sta(mvm, skb) == 0))
+			return;
+	}
+
+	ieee80211_free_txskb(mvm->hw, skb);
+}
+
 static void iwl_mvm_mac_tx(struct ieee80211_hw *hw,
 			   struct ieee80211_tx_control *control,
 			   struct sk_buff *skb)
@@ -797,14 +812,7 @@ static void iwl_mvm_mac_tx(struct ieee80211_hw *hw,
 		}
 	}
 
-	if (sta) {
-		if (iwl_mvm_tx_skb(mvm, skb, sta))
-			goto drop;
-		return;
-	}
-
-	if (iwl_mvm_tx_skb_non_sta(mvm, skb))
-		goto drop;
+	iwl_mvm_tx_skb(mvm, skb, sta);
 	return;
  drop:
 	ieee80211_free_txskb(hw, skb);
@@ -854,10 +862,7 @@ void iwl_mvm_mac_itxq_xmit(struct ieee80211_hw *hw, struct ieee80211_txq *txq)
 				break;
 			}
 
-			if (!txq->sta)
-				iwl_mvm_tx_skb_non_sta(mvm, skb);
-			else
-				iwl_mvm_tx_skb(mvm, skb, txq->sta);
+			iwl_mvm_tx_skb(mvm, skb, txq->sta);
 		}
 	} while (atomic_dec_return(&mvmtxq->tx_request));
 	rcu_read_unlock();
@@ -4771,6 +4776,125 @@ static int iwl_mvm_mac_get_survey(struct ieee80211_hw *hw, int idx,
 	return ret;
 }
 
+static void iwl_mvm_set_sta_rate(u32 rate_n_flags, struct rate_info *rinfo)
+{
+	switch (rate_n_flags & RATE_MCS_CHAN_WIDTH_MSK) {
+	case RATE_MCS_CHAN_WIDTH_20:
+		rinfo->bw = RATE_INFO_BW_20;
+		break;
+	case RATE_MCS_CHAN_WIDTH_40:
+		rinfo->bw = RATE_INFO_BW_40;
+		break;
+	case RATE_MCS_CHAN_WIDTH_80:
+		rinfo->bw = RATE_INFO_BW_80;
+		break;
+	case RATE_MCS_CHAN_WIDTH_160:
+		rinfo->bw = RATE_INFO_BW_160;
+		break;
+	}
+
+	if (rate_n_flags & RATE_MCS_HT_MSK) {
+		rinfo->flags |= RATE_INFO_FLAGS_MCS;
+		rinfo->mcs = u32_get_bits(rate_n_flags, RATE_HT_MCS_INDEX_MSK);
+		rinfo->nss = u32_get_bits(rate_n_flags,
+					  RATE_HT_MCS_NSS_MSK) + 1;
+		if (rate_n_flags & RATE_MCS_SGI_MSK)
+			rinfo->flags |= RATE_INFO_FLAGS_SHORT_GI;
+	} else if (rate_n_flags & RATE_MCS_VHT_MSK) {
+		rinfo->flags |= RATE_INFO_FLAGS_VHT_MCS;
+		rinfo->mcs = u32_get_bits(rate_n_flags,
+					  RATE_VHT_MCS_RATE_CODE_MSK);
+		rinfo->nss = u32_get_bits(rate_n_flags,
+					  RATE_VHT_MCS_NSS_MSK) + 1;
+		if (rate_n_flags & RATE_MCS_SGI_MSK)
+			rinfo->flags |= RATE_INFO_FLAGS_SHORT_GI;
+	} else if (rate_n_flags & RATE_MCS_HE_MSK) {
+		u32 gi_ltf = u32_get_bits(rate_n_flags,
+					  RATE_MCS_HE_GI_LTF_MSK);
+
+		rinfo->flags |= RATE_INFO_FLAGS_HE_MCS;
+		rinfo->mcs = u32_get_bits(rate_n_flags,
+					  RATE_VHT_MCS_RATE_CODE_MSK);
+		rinfo->nss = u32_get_bits(rate_n_flags,
+					  RATE_VHT_MCS_NSS_MSK) + 1;
+
+		if (rate_n_flags & RATE_MCS_HE_106T_MSK) {
+			rinfo->bw = RATE_INFO_BW_HE_RU;
+			rinfo->he_ru_alloc = NL80211_RATE_INFO_HE_RU_ALLOC_106;
+		}
+
+		switch (rate_n_flags & RATE_MCS_HE_TYPE_MSK) {
+		case RATE_MCS_HE_TYPE_SU:
+		case RATE_MCS_HE_TYPE_EXT_SU:
+			if (gi_ltf == 0 || gi_ltf == 1)
+				rinfo->he_gi = NL80211_RATE_INFO_HE_GI_0_8;
+			else if (gi_ltf == 2)
+				rinfo->he_gi = NL80211_RATE_INFO_HE_GI_1_6;
+			else if (rate_n_flags & RATE_MCS_SGI_MSK)
+				rinfo->he_gi = NL80211_RATE_INFO_HE_GI_0_8;
+			else
+				rinfo->he_gi = NL80211_RATE_INFO_HE_GI_3_2;
+			break;
+		case RATE_MCS_HE_TYPE_MU:
+			if (gi_ltf == 0 || gi_ltf == 1)
+				rinfo->he_gi = NL80211_RATE_INFO_HE_GI_0_8;
+			else if (gi_ltf == 2)
+				rinfo->he_gi = NL80211_RATE_INFO_HE_GI_1_6;
+			else
+				rinfo->he_gi = NL80211_RATE_INFO_HE_GI_3_2;
+			break;
+		case RATE_MCS_HE_TYPE_TRIG:
+			if (gi_ltf == 0 || gi_ltf == 1)
+				rinfo->he_gi = NL80211_RATE_INFO_HE_GI_1_6;
+			else
+				rinfo->he_gi = NL80211_RATE_INFO_HE_GI_3_2;
+			break;
+		}
+
+		if (rate_n_flags & RATE_HE_DUAL_CARRIER_MODE_MSK)
+			rinfo->he_dcm = 1;
+	} else {
+		switch (u32_get_bits(rate_n_flags, RATE_LEGACY_RATE_MSK)) {
+		case IWL_RATE_1M_PLCP:
+			rinfo->legacy = 10;
+			break;
+		case IWL_RATE_2M_PLCP:
+			rinfo->legacy = 20;
+			break;
+		case IWL_RATE_5M_PLCP:
+			rinfo->legacy = 55;
+			break;
+		case IWL_RATE_11M_PLCP:
+			rinfo->legacy = 110;
+			break;
+		case IWL_RATE_6M_PLCP:
+			rinfo->legacy = 60;
+			break;
+		case IWL_RATE_9M_PLCP:
+			rinfo->legacy = 90;
+			break;
+		case IWL_RATE_12M_PLCP:
+			rinfo->legacy = 120;
+			break;
+		case IWL_RATE_18M_PLCP:
+			rinfo->legacy = 180;
+			break;
+		case IWL_RATE_24M_PLCP:
+			rinfo->legacy = 240;
+			break;
+		case IWL_RATE_36M_PLCP:
+			rinfo->legacy = 360;
+			break;
+		case IWL_RATE_48M_PLCP:
+			rinfo->legacy = 480;
+			break;
+		case IWL_RATE_54M_PLCP:
+			rinfo->legacy = 540;
+			break;
+		}
+	}
+}
+
 static void iwl_mvm_mac_sta_statistics(struct ieee80211_hw *hw,
 				       struct ieee80211_vif *vif,
 				       struct ieee80211_sta *sta,
@@ -4785,6 +4909,13 @@ static void iwl_mvm_mac_sta_statistics(struct ieee80211_hw *hw,
 		sinfo->filled |= BIT_ULL(NL80211_STA_INFO_SIGNAL_AVG);
 	}
 
+	if (iwl_mvm_has_tlc_offload(mvm)) {
+		struct iwl_lq_sta_rs_fw *lq_sta = &mvmsta->lq_sta.rs_fw;
+
+		iwl_mvm_set_sta_rate(lq_sta->last_rate_n_flags, &sinfo->txrate);
+		sinfo->filled |= BIT_ULL(NL80211_STA_INFO_TX_BITRATE);
+	}
+
 	/* if beacon filtering isn't on mac80211 does it anyway */
 	if (!(vif->driver_flags & IEEE80211_VIF_BEACON_FILTER))
 		return;

drivers/net/wireless/intel/iwlwifi/mvm/mvm.h

@@ -1298,9 +1298,6 @@ static inline bool iwl_mvm_is_lar_supported(struct iwl_mvm *mvm)
 	bool tlv_lar = fw_has_capa(&mvm->fw->ucode_capa,
 				   IWL_UCODE_TLV_CAPA_LAR_SUPPORT);
 
-	if (iwlwifi_mod_params.lar_disable)
-		return false;
-
 	/*
 	 * Enable LAR only if it is supported by the FW (TLV) &&
 	 * enabled in the NVM
@@ -1508,8 +1505,8 @@ int __must_check iwl_mvm_send_cmd_status(struct iwl_mvm *mvm,
 int __must_check iwl_mvm_send_cmd_pdu_status(struct iwl_mvm *mvm, u32 id,
 					     u16 len, const void *data,
 					     u32 *status);
-int iwl_mvm_tx_skb(struct iwl_mvm *mvm, struct sk_buff *skb,
-		   struct ieee80211_sta *sta);
+int iwl_mvm_tx_skb_sta(struct iwl_mvm *mvm, struct sk_buff *skb,
+		       struct ieee80211_sta *sta);
 int iwl_mvm_tx_skb_non_sta(struct iwl_mvm *mvm, struct sk_buff *skb);
 void iwl_mvm_set_tx_cmd(struct iwl_mvm *mvm, struct sk_buff *skb,
 			struct iwl_tx_cmd *tx_cmd,

drivers/net/wireless/intel/iwlwifi/mvm/nvm.c

@@ -277,11 +277,10 @@ iwl_parse_nvm_sections(struct iwl_mvm *mvm)
 	struct iwl_nvm_section *sections = mvm->nvm_sections;
 	const __be16 *hw;
 	const __le16 *sw, *calib, *regulatory, *mac_override, *phy_sku;
-	bool lar_enabled;
 	int regulatory_type;
 
 	/* Checking for required sections */
-	if (mvm->trans->cfg->nvm_type != IWL_NVM_EXT) {
+	if (mvm->trans->cfg->nvm_type == IWL_NVM) {
 		if (!mvm->nvm_sections[NVM_SECTION_TYPE_SW].data ||
 		    !mvm->nvm_sections[mvm->cfg->nvm_hw_section_num].data) {
 			IWL_ERR(mvm, "Can't parse empty OTP/NVM sections\n");
@@ -327,14 +326,9 @@ iwl_parse_nvm_sections(struct iwl_mvm *mvm)
 		(const __le16 *)sections[NVM_SECTION_TYPE_REGULATORY_SDP].data :
 		(const __le16 *)sections[NVM_SECTION_TYPE_REGULATORY].data;
 
-	lar_enabled = !iwlwifi_mod_params.lar_disable &&
-		      fw_has_capa(&mvm->fw->ucode_capa,
-				  IWL_UCODE_TLV_CAPA_LAR_SUPPORT);
-
-	return iwl_parse_nvm_data(mvm->trans, mvm->cfg, hw, sw, calib,
+	return iwl_parse_nvm_data(mvm->trans, mvm->cfg, mvm->fw, hw, sw, calib,
 				  regulatory, mac_override, phy_sku,
-				  mvm->fw->valid_tx_ant, mvm->fw->valid_rx_ant,
-				  lar_enabled);
+				  mvm->fw->valid_tx_ant, mvm->fw->valid_rx_ant);
 }
 
 /* Loads the NVM data stored in mvm->nvm_sections into the NIC */

drivers/net/wireless/intel/iwlwifi/mvm/scan.c

@@ -1213,7 +1213,7 @@ static int iwl_mvm_legacy_config_scan(struct iwl_mvm *mvm)
 		cmd_size = sizeof(struct iwl_scan_config_v2);
 	else
 		cmd_size = sizeof(struct iwl_scan_config_v1);
-	cmd_size += num_channels;
+	cmd_size += mvm->fw->ucode_capa.n_scan_channels;
 
 	cfg = kzalloc(cmd_size, GFP_KERNEL);
 	if (!cfg)

drivers/net/wireless/intel/iwlwifi/mvm/tx.c

@@ -490,13 +490,13 @@ static void iwl_mvm_set_tx_cmd_crypto(struct iwl_mvm *mvm,
 /*
  * Allocates and sets the Tx cmd the driver data pointers in the skb
  */
-static struct iwl_device_cmd *
+static struct iwl_device_tx_cmd *
 iwl_mvm_set_tx_params(struct iwl_mvm *mvm, struct sk_buff *skb,
 		      struct ieee80211_tx_info *info, int hdrlen,
 		      struct ieee80211_sta *sta, u8 sta_id)
 {
 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
-	struct iwl_device_cmd *dev_cmd;
+	struct iwl_device_tx_cmd *dev_cmd;
 	struct iwl_tx_cmd *tx_cmd;
 
 	dev_cmd = iwl_trans_alloc_tx_cmd(mvm->trans);
@@ -504,11 +504,6 @@ iwl_mvm_set_tx_params(struct iwl_mvm *mvm, struct sk_buff *skb,
 	if (unlikely(!dev_cmd))
 		return NULL;
 
-	/* Make sure we zero enough of dev_cmd */
-	BUILD_BUG_ON(sizeof(struct iwl_tx_cmd_gen2) > sizeof(*tx_cmd));
-	BUILD_BUG_ON(sizeof(struct iwl_tx_cmd_gen3) > sizeof(*tx_cmd));
-
-	memset(dev_cmd, 0, sizeof(dev_cmd->hdr) + sizeof(*tx_cmd));
 	dev_cmd->hdr.cmd = TX_CMD;
 
 	if (iwl_mvm_has_new_tx_api(mvm)) {
@@ -597,7 +592,7 @@ iwl_mvm_set_tx_params(struct iwl_mvm *mvm, struct sk_buff *skb,
 }
 
 static void iwl_mvm_skb_prepare_status(struct sk_buff *skb,
-				       struct iwl_device_cmd *cmd)
+				       struct iwl_device_tx_cmd *cmd)
 {
 	struct ieee80211_tx_info *skb_info = IEEE80211_SKB_CB(skb);
 
@@ -716,7 +711,7 @@ int iwl_mvm_tx_skb_non_sta(struct iwl_mvm *mvm, struct sk_buff *skb)
 {
 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
 	struct ieee80211_tx_info info;
-	struct iwl_device_cmd *dev_cmd;
+	struct iwl_device_tx_cmd *dev_cmd;
 	u8 sta_id;
 	int hdrlen = ieee80211_hdrlen(hdr->frame_control);
 	__le16 fc = hdr->frame_control;
@@ -1078,7 +1073,7 @@ static int iwl_mvm_tx_mpdu(struct iwl_mvm *mvm, struct sk_buff *skb,
 {
 	struct ieee80211_hdr *hdr = (struct ieee80211_hdr *)skb->data;
 	struct iwl_mvm_sta *mvmsta;
-	struct iwl_device_cmd *dev_cmd;
+	struct iwl_device_tx_cmd *dev_cmd;
 	__le16 fc;
 	u16 seq_number = 0;
 	u8 tid = IWL_MAX_TID_COUNT;
@@ -1154,7 +1149,7 @@ static int iwl_mvm_tx_mpdu(struct iwl_mvm *mvm, struct sk_buff *skb,
 	if (WARN_ONCE(txq_id == IWL_MVM_INVALID_QUEUE, "Invalid TXQ id")) {
 		iwl_trans_free_tx_cmd(mvm->trans, dev_cmd);
 		spin_unlock(&mvmsta->lock);
-		return 0;
+		return -1;
 	}
 
 	if (!iwl_mvm_has_new_tx_api(mvm)) {
@@ -1206,8 +1201,8 @@ static int iwl_mvm_tx_mpdu(struct iwl_mvm *mvm, struct sk_buff *skb,
 	return -1;
 }
 
-int iwl_mvm_tx_skb(struct iwl_mvm *mvm, struct sk_buff *skb,
-		   struct ieee80211_sta *sta)
+int iwl_mvm_tx_skb_sta(struct iwl_mvm *mvm, struct sk_buff *skb,
+		       struct ieee80211_sta *sta)
 {
 	struct iwl_mvm_sta *mvmsta = iwl_mvm_sta_from_mac80211(sta);
 	struct ieee80211_tx_info info;

drivers/net/wireless/intel/iwlwifi/pcie/ctxt-info.c

@@ -57,6 +57,42 @@
 #include "internal.h"
 #include "iwl-prph.h"
 
+static void *_iwl_pcie_ctxt_info_dma_alloc_coherent(struct iwl_trans *trans,
+						    size_t size,
+						    dma_addr_t *phys,
+						    int depth)
+{
+	void *result;
+
+	if (WARN(depth > 2,
+		 "failed to allocate DMA memory not crossing 2^32 boundary"))
+		return NULL;
+
+	result = dma_alloc_coherent(trans->dev, size, phys, GFP_KERNEL);
+
+	if (!result)
+		return NULL;
+
+	if (unlikely(iwl_pcie_crosses_4g_boundary(*phys, size))) {
+		void *old = result;
+		dma_addr_t oldphys = *phys;
+
+		result = _iwl_pcie_ctxt_info_dma_alloc_coherent(trans, size,
+								phys,
+								depth + 1);
+		dma_free_coherent(trans->dev, size, old, oldphys);
+	}
+
+	return result;
+}
+
+static void *iwl_pcie_ctxt_info_dma_alloc_coherent(struct iwl_trans *trans,
+						   size_t size,
+						   dma_addr_t *phys)
+{
+	return _iwl_pcie_ctxt_info_dma_alloc_coherent(trans, size, phys, 0);
+}
+
 void iwl_pcie_ctxt_info_free_paging(struct iwl_trans *trans)
 {
 	struct iwl_self_init_dram *dram = &trans->init_dram;
@@ -161,14 +197,17 @@ int iwl_pcie_ctxt_info_init(struct iwl_trans *trans,
 	struct iwl_context_info *ctxt_info;
 	struct iwl_context_info_rbd_cfg *rx_cfg;
 	u32 control_flags = 0, rb_size;
+	dma_addr_t phys;
 	int ret;
 
-	ctxt_info = dma_alloc_coherent(trans->dev, sizeof(*ctxt_info),
-				       &trans_pcie->ctxt_info_dma_addr,
-				       GFP_KERNEL);
+	ctxt_info = iwl_pcie_ctxt_info_dma_alloc_coherent(trans,
+							  sizeof(*ctxt_info),
+							  &phys);
 	if (!ctxt_info)
 		return -ENOMEM;
 
+	trans_pcie->ctxt_info_dma_addr = phys;
+
 	ctxt_info->version.version = 0;
 	ctxt_info->version.mac_id =
 		cpu_to_le16((u16)iwl_read32(trans, CSR_HW_REV));

drivers/net/wireless/intel/iwlwifi/pcie/internal.h

@@ -305,7 +305,7 @@ struct iwl_cmd_meta {
 #define IWL_FIRST_TB_SIZE_ALIGN ALIGN(IWL_FIRST_TB_SIZE, 64)
 
 struct iwl_pcie_txq_entry {
-	struct iwl_device_cmd *cmd;
+	void *cmd;
 	struct sk_buff *skb;
 	/* buffer to free after command completes */
 	const void *free_buf;
@@ -672,6 +672,16 @@ void iwl_pcie_disable_ict(struct iwl_trans *trans);
 /*****************************************************
 * TX / HCMD
 ******************************************************/
+/*
+ * 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_pcie_crosses_4g_boundary(u64 phys, u16 len)
+{
+	return upper_32_bits(phys) != upper_32_bits(phys + len);
+}
+
 int iwl_pcie_tx_init(struct iwl_trans *trans);
 int iwl_pcie_gen2_tx_init(struct iwl_trans *trans, int txq_id,
 			  int queue_size);
@@ -688,7 +698,7 @@ void iwl_trans_pcie_txq_set_shared_mode(struct iwl_trans *trans, u32 txq_id,
 void iwl_trans_pcie_log_scd_error(struct iwl_trans *trans,
 				  struct iwl_txq *txq);
 int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
-		      struct iwl_device_cmd *dev_cmd, int txq_id);
+		      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_cmdq_reclaim(struct iwl_trans *trans, int txq_id, int idx);
@@ -1082,7 +1092,8 @@ void iwl_pcie_apply_destination(struct iwl_trans *trans);
 void iwl_pcie_free_tso_page(struct iwl_trans_pcie *trans_pcie,
 			    struct sk_buff *skb);
 #ifdef CONFIG_INET
-struct iwl_tso_hdr_page *get_page_hdr(struct iwl_trans *trans, size_t len);
+struct iwl_tso_hdr_page *get_page_hdr(struct iwl_trans *trans, size_t len,
+				      struct sk_buff *skb);
 #endif
 
 /* common functions that are used by gen3 transport */
@@ -1106,7 +1117,7 @@ int iwl_trans_pcie_dyn_txq_alloc(struct iwl_trans *trans,
 				 unsigned int timeout);
 void iwl_trans_pcie_dyn_txq_free(struct iwl_trans *trans, int queue);
 int iwl_trans_pcie_gen2_tx(struct iwl_trans *trans, struct sk_buff *skb,
-			   struct iwl_device_cmd *dev_cmd, int txq_id);
+			   struct iwl_device_tx_cmd *dev_cmd, int txq_id);
 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);

drivers/net/wireless/intel/iwlwifi/pcie/trans.c

@@ -79,6 +79,7 @@
 #include "iwl-agn-hw.h"
 #include "fw/error-dump.h"
 #include "fw/dbg.h"
+#include "fw/api/tx.h"
 #include "internal.h"
 #include "iwl-fh.h"
 
@@ -301,18 +302,13 @@ void iwl_pcie_apm_config(struct iwl_trans *trans)
 	u16 cap;
 
 	/*
-	 * HW bug W/A for instability in PCIe bus L0S->L1 transition.
-	 * Check if BIOS (or OS) enabled L1-ASPM on this device.
-	 * If so (likely), disable L0S, so device moves directly L0->L1;
-	 *    costs negligible amount of power savings.
-	 * If not (unlikely), enable L0S, so there is at least some
-	 *    power savings, even without L1.
+	 * L0S states have been found to be unstable with our devices
+	 * and in newer hardware they are not officially supported at
+	 * all, so we must always set the L0S_DISABLED bit.
 	 */
+	iwl_set_bit(trans, CSR_GIO_REG, CSR_GIO_REG_VAL_L0S_DISABLED);
+
 	pcie_capability_read_word(trans_pcie->pci_dev, PCI_EXP_LNKCTL, &lctl);
-	if (lctl & PCI_EXP_LNKCTL_ASPM_L1)
-		iwl_set_bit(trans, CSR_GIO_REG, CSR_GIO_REG_VAL_L0S_ENABLED);
-	else
-		iwl_clear_bit(trans, CSR_GIO_REG, CSR_GIO_REG_VAL_L0S_ENABLED);
 	trans->pm_support = !(lctl & PCI_EXP_LNKCTL_ASPM_L0S);
 
 	pcie_capability_read_word(trans_pcie->pci_dev, PCI_EXP_DEVCTL2, &cap);
@@ -3460,19 +3456,34 @@ struct iwl_trans *iwl_trans_pcie_alloc(struct pci_dev *pdev,
 {
 	struct iwl_trans_pcie *trans_pcie;
 	struct iwl_trans *trans;
-	int ret, addr_size;
+	int ret, addr_size, txcmd_size, txcmd_align;
+	const struct iwl_trans_ops *ops = &trans_ops_pcie_gen2;
+
+	if (!cfg_trans->gen2) {
+		ops = &trans_ops_pcie;
+		txcmd_size = sizeof(struct iwl_tx_cmd);
+		txcmd_align = sizeof(void *);
+	} else if (cfg_trans->device_family < IWL_DEVICE_FAMILY_AX210) {
+		txcmd_size = sizeof(struct iwl_tx_cmd_gen2);
+		txcmd_align = 64;
+	} else {
+		txcmd_size = sizeof(struct iwl_tx_cmd_gen3);
+		txcmd_align = 128;
+	}
+
+	txcmd_size += sizeof(struct iwl_cmd_header);
+	txcmd_size += 36; /* biggest possible 802.11 header */
+
+	/* Ensure device TX cmd cannot reach/cross a page boundary in gen2 */
+	if (WARN_ON(cfg_trans->gen2 && txcmd_size >= txcmd_align))
+		return ERR_PTR(-EINVAL);
 
 	ret = pcim_enable_device(pdev);
 	if (ret)
 		return ERR_PTR(ret);
 
-	if (cfg_trans->gen2)
-		trans = iwl_trans_alloc(sizeof(struct iwl_trans_pcie),
-					&pdev->dev, &trans_ops_pcie_gen2);
-	else
-		trans = iwl_trans_alloc(sizeof(struct iwl_trans_pcie),
-					&pdev->dev, &trans_ops_pcie);
-
+	trans = iwl_trans_alloc(sizeof(struct iwl_trans_pcie), &pdev->dev, ops,
+				txcmd_size, txcmd_align);
 	if (!trans)
 		return ERR_PTR(-ENOMEM);
 

drivers/net/wireless/intel/iwlwifi/pcie/tx-gen2.c

@@ -221,6 +221,17 @@ static int iwl_pcie_gen2_set_tb(struct iwl_trans *trans,
 	int idx = iwl_pcie_gen2_get_num_tbs(trans, 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_pcie_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];
@@ -240,13 +251,114 @@ static int iwl_pcie_gen2_set_tb(struct iwl_trans *trans,
 	return idx;
 }
 
+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 page **page_ptr;
+	struct page *ret;
+
+	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
+
+	ret = alloc_page(GFP_ATOMIC);
+	if (!ret)
+		return NULL;
+
+	/* set the chaining pointer to the previous page if there */
+	*(void **)(page_address(ret) + PAGE_SIZE - sizeof(void *)) = *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_pcie_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)
+{
+	dma_addr_t oldphys = phys;
+	struct page *page;
+	int ret;
+
+	if (unlikely(dma_mapping_error(trans->dev, phys)))
+		return -ENOMEM;
+
+	if (likely(!iwl_pcie_crosses_4g_boundary(phys, len))) {
+		ret = iwl_pcie_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 > PAGE_SIZE - sizeof(void *))) {
+		ret = -ENOBUFS;
+		goto unmap;
+	}
+
+	page = get_workaround_page(trans, skb);
+	if (!page) {
+		ret = -ENOMEM;
+		goto unmap;
+	}
+
+	memcpy(page_address(page), virt, len);
+
+	phys = dma_map_single(trans->dev, page_address(page), len,
+			      DMA_TO_DEVICE);
+	if (unlikely(dma_mapping_error(trans->dev, phys)))
+		return -ENOMEM;
+	ret = iwl_pcie_gen2_set_tb(trans, tfd, phys, len);
+	if (ret < 0) {
+		/* unmap the new allocation as single */
+		oldphys = phys;
+		meta = NULL;
+		goto unmap;
+	}
+	IWL_WARN(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 (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_pcie_gen2_build_amsdu(struct iwl_trans *trans,
 				     struct sk_buff *skb,
 				     struct iwl_tfh_tfd *tfd, int start_len,
-				     u8 hdr_len, struct iwl_device_cmd *dev_cmd)
+				     u8 hdr_len,
+				     struct iwl_device_tx_cmd *dev_cmd)
 {
 #ifdef CONFIG_INET
-	struct iwl_trans_pcie *trans_pcie = IWL_TRANS_GET_PCIE_TRANS(trans);
 	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;
@@ -254,7 +366,6 @@ static int iwl_pcie_gen2_build_amsdu(struct iwl_trans *trans,
 	u16 length, amsdu_pad;
 	u8 *start_hdr;
 	struct iwl_tso_hdr_page *hdr_page;
-	struct page **page_ptr;
 	struct tso_t tso;
 
 	trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd),
@@ -270,14 +381,11 @@ static int iwl_pcie_gen2_build_amsdu(struct iwl_trans *trans,
 		(3 + snap_ip_tcp_hdrlen + sizeof(struct ethhdr));
 
 	/* Our device supports 9 segments at most, it will fit in 1 page */
-	hdr_page = get_page_hdr(trans, hdr_room);
+	hdr_page = get_page_hdr(trans, hdr_room, skb);
 	if (!hdr_page)
 		return -ENOMEM;
 
-	get_page(hdr_page->page);
 	start_hdr = hdr_page->pos;
-	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
-	*page_ptr = hdr_page->page;
 
 	/*
 	 * Pull the ieee80211 header to be able to use TSO core,
@@ -332,6 +440,11 @@ static int iwl_pcie_gen2_build_amsdu(struct iwl_trans *trans,
 			dev_kfree_skb(csum_skb);
 			goto out_err;
 		}
+		/*
+		 * 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_pcie_gen2_set_tb(trans, tfd, tb_phys, tb_len);
 		trace_iwlwifi_dev_tx_tb(trans->dev, skb, start_hdr,
 					tb_phys, tb_len);
@@ -343,16 +456,18 @@ static int iwl_pcie_gen2_build_amsdu(struct iwl_trans *trans,
 
 		/* put the payload */
 		while (data_left) {
+			int ret;
+
 			tb_len = min_t(unsigned int, tso.size, data_left);
 			tb_phys = dma_map_single(trans->dev, tso.data,
 						 tb_len, DMA_TO_DEVICE);
-			if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
+			ret = iwl_pcie_gen2_set_tb_with_wa(trans, skb, tfd,
+							   tb_phys, tso.data,
+							   tb_len, NULL);
+			if (ret) {
 				dev_kfree_skb(csum_skb);
 				goto out_err;
 			}
-			iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb_len);
-			trace_iwlwifi_dev_tx_tb(trans->dev, skb, tso.data,
-						tb_phys, tb_len);
 
 			data_left -= tb_len;
 			tso_build_data(skb, &tso, tb_len);
@@ -372,7 +487,7 @@ static int iwl_pcie_gen2_build_amsdu(struct iwl_trans *trans,
 static struct
 iwl_tfh_tfd *iwl_pcie_gen2_build_tx_amsdu(struct iwl_trans *trans,
 					  struct iwl_txq *txq,
-					  struct iwl_device_cmd *dev_cmd,
+					  struct iwl_device_tx_cmd *dev_cmd,
 					  struct sk_buff *skb,
 					  struct iwl_cmd_meta *out_meta,
 					  int hdr_len,
@@ -386,6 +501,11 @@ iwl_tfh_tfd *iwl_pcie_gen2_build_tx_amsdu(struct iwl_trans *trans,
 
 	tb_phys = iwl_pcie_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_pcie_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE);
 
 	/*
@@ -404,6 +524,10 @@ iwl_tfh_tfd *iwl_pcie_gen2_build_tx_amsdu(struct iwl_trans *trans,
 	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_pcie_gen2_set_tb(trans, tfd, tb_phys, len);
 
 	if (iwl_pcie_gen2_build_amsdu(trans, skb, tfd,
@@ -430,24 +554,19 @@ static int iwl_pcie_gen2_tx_add_frags(struct iwl_trans *trans,
 	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;
-		int tb_idx;
+		unsigned int fragsz = skb_frag_size(frag);
+		int ret;
 
-		if (!skb_frag_size(frag))
+		if (!fragsz)
 			continue;
 
 		tb_phys = skb_frag_dma_map(trans->dev, frag, 0,
-					   skb_frag_size(frag), DMA_TO_DEVICE);
-
-		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
-			return -ENOMEM;
-		tb_idx = iwl_pcie_gen2_set_tb(trans, tfd, tb_phys,
-					      skb_frag_size(frag));
-		trace_iwlwifi_dev_tx_tb(trans->dev, skb, skb_frag_address(frag),
-					tb_phys, skb_frag_size(frag));
-		if (tb_idx < 0)
-			return tb_idx;
-
-		out_meta->tbs |= BIT(tb_idx);
+					   fragsz, DMA_TO_DEVICE);
+		ret = iwl_pcie_gen2_set_tb_with_wa(trans, skb, tfd, tb_phys,
+						   skb_frag_address(frag),
+						   fragsz, out_meta);
+		if (ret)
+			return ret;
 	}
 
 	return 0;
@@ -456,7 +575,7 @@ static int iwl_pcie_gen2_tx_add_frags(struct iwl_trans *trans,
 static struct
 iwl_tfh_tfd *iwl_pcie_gen2_build_tx(struct iwl_trans *trans,
 				    struct iwl_txq *txq,
-				    struct iwl_device_cmd *dev_cmd,
+				    struct iwl_device_tx_cmd *dev_cmd,
 				    struct sk_buff *skb,
 				    struct iwl_cmd_meta *out_meta,
 				    int hdr_len,
@@ -475,6 +594,11 @@ iwl_tfh_tfd *iwl_pcie_gen2_build_tx(struct iwl_trans *trans,
 	/* 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_pcie_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE);
 
 	/*
@@ -496,6 +620,10 @@ iwl_tfh_tfd *iwl_pcie_gen2_build_tx(struct iwl_trans *trans,
 	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_pcie_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);
@@ -504,26 +632,30 @@ iwl_tfh_tfd *iwl_pcie_gen2_build_tx(struct iwl_trans *trans,
 	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);
-		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
+		ret = iwl_pcie_gen2_set_tb_with_wa(trans, skb, tfd, tb_phys,
+						   skb->data + hdr_len, tb2_len,
+						   NULL);
+		if (ret)
 			goto out_err;
-		iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb2_len);
-		trace_iwlwifi_dev_tx_tb(trans->dev, skb, skb->data + hdr_len,
-					tb_phys, tb2_len);
 	}
 
 	if (iwl_pcie_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);
-		if (unlikely(dma_mapping_error(trans->dev, tb_phys)))
+		ret = iwl_pcie_gen2_set_tb_with_wa(trans, skb, tfd, tb_phys,
+						   frag->data,
+						   skb_headlen(frag), NULL);
+		if (ret)
 			goto out_err;
-		iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, skb_headlen(frag));
-		trace_iwlwifi_dev_tx_tb(trans->dev, skb, frag->data,
-					tb_phys, skb_headlen(frag));
 		if (iwl_pcie_gen2_tx_add_frags(trans, frag, tfd, out_meta))
 			goto out_err;
 	}
@@ -538,7 +670,7 @@ iwl_tfh_tfd *iwl_pcie_gen2_build_tx(struct iwl_trans *trans,
 static
 struct iwl_tfh_tfd *iwl_pcie_gen2_build_tfd(struct iwl_trans *trans,
 					    struct iwl_txq *txq,
-					    struct iwl_device_cmd *dev_cmd,
+					    struct iwl_device_tx_cmd *dev_cmd,
 					    struct sk_buff *skb,
 					    struct iwl_cmd_meta *out_meta)
 {
@@ -578,7 +710,7 @@ struct iwl_tfh_tfd *iwl_pcie_gen2_build_tfd(struct iwl_trans *trans,
 }
 
 int iwl_trans_pcie_gen2_tx(struct iwl_trans *trans, struct sk_buff *skb,
-			   struct iwl_device_cmd *dev_cmd, int txq_id)
+			   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;
@@ -603,7 +735,7 @@ int iwl_trans_pcie_gen2_tx(struct iwl_trans *trans, struct sk_buff *skb,
 
 		/* don't put the packet on the ring, if there is no room */
 		if (unlikely(iwl_queue_space(trans, txq) < 3)) {
-			struct iwl_device_cmd **dev_cmd_ptr;
+			struct iwl_device_tx_cmd **dev_cmd_ptr;
 
 			dev_cmd_ptr = (void *)((u8 *)skb->cb +
 					       trans_pcie->dev_cmd_offs);

drivers/net/wireless/intel/iwlwifi/pcie/tx.c

@@ -213,8 +213,8 @@ static void iwl_pcie_txq_update_byte_cnt_tbl(struct iwl_trans *trans,
 	u8 sec_ctl = 0;
 	u16 len = byte_cnt + IWL_TX_CRC_SIZE + IWL_TX_DELIMITER_SIZE;
 	__le16 bc_ent;
-	struct iwl_tx_cmd *tx_cmd =
-		(void *)txq->entries[txq->write_ptr].cmd->payload;
+	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->scd_bc_tbls.addr;
@@ -257,8 +257,8 @@ static void iwl_pcie_txq_inval_byte_cnt_tbl(struct iwl_trans *trans,
 	int read_ptr = txq->read_ptr;
 	u8 sta_id = 0;
 	__le16 bc_ent;
-	struct iwl_tx_cmd *tx_cmd =
-		(void *)txq->entries[read_ptr].cmd->payload;
+	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);
 
@@ -624,12 +624,18 @@ void iwl_pcie_free_tso_page(struct iwl_trans_pcie *trans_pcie,
 			    struct sk_buff *skb)
 {
 	struct page **page_ptr;
+	struct page *next;
 
 	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
+	next = *page_ptr;
+	*page_ptr = NULL;
 
-	if (*page_ptr) {
-		__free_page(*page_ptr);
-		*page_ptr = NULL;
+	while (next) {
+		struct page *tmp = next;
+
+		next = *(void **)(page_address(next) + PAGE_SIZE -
+				  sizeof(void *));
+		__free_page(tmp);
 	}
 }
 
@@ -1196,7 +1202,7 @@ void iwl_trans_pcie_reclaim(struct iwl_trans *trans, int txq_id, int ssn,
 
 		while (!skb_queue_empty(&overflow_skbs)) {
 			struct sk_buff *skb = __skb_dequeue(&overflow_skbs);
-			struct iwl_device_cmd *dev_cmd_ptr;
+			struct iwl_device_tx_cmd *dev_cmd_ptr;
 
 			dev_cmd_ptr = *(void **)((u8 *)skb->cb +
 						 trans_pcie->dev_cmd_offs);
@@ -2052,17 +2058,34 @@ static int iwl_fill_data_tbs(struct iwl_trans *trans, struct sk_buff *skb,
 }
 
 #ifdef CONFIG_INET
-struct iwl_tso_hdr_page *get_page_hdr(struct iwl_trans *trans, size_t len)
+struct iwl_tso_hdr_page *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->tso_hdr_page);
+	struct page **page_ptr;
+
+	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
+
+	if (WARN_ON(*page_ptr))
+		return NULL;
 
 	if (!p->page)
 		goto alloc;
 
-	/* enough room on this page */
-	if (p->pos + len < (u8 *)page_address(p->page) + PAGE_SIZE)
-		return p;
+	/*
+	 * 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 (p->pos + len < (u8 *)page_address(p->page) + PAGE_SIZE -
+			   sizeof(void *))
+		goto out;
 
 	/* We don't have enough room on this page, get a new one. */
 	__free_page(p->page);
@@ -2072,6 +2095,11 @@ struct iwl_tso_hdr_page *get_page_hdr(struct iwl_trans *trans, size_t len)
 	if (!p->page)
 		return NULL;
 	p->pos = page_address(p->page);
+	/* set the chaining pointer to NULL */
+	*(void **)(page_address(p->page) + PAGE_SIZE - sizeof(void *)) = NULL;
+out:
+	*page_ptr = p->page;
+	get_page(p->page);
 	return p;
 }
 
@@ -2097,7 +2125,8 @@ static void iwl_compute_pseudo_hdr_csum(void *iph, struct tcphdr *tcph,
 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_cmd *dev_cmd, u16 tb1_len)
+				   struct iwl_device_tx_cmd *dev_cmd,
+				   u16 tb1_len)
 {
 	struct iwl_tx_cmd *tx_cmd = (void *)dev_cmd->payload;
 	struct iwl_trans_pcie *trans_pcie = txq->trans_pcie;
@@ -2107,7 +2136,6 @@ static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
 	u16 length, iv_len, amsdu_pad;
 	u8 *start_hdr;
 	struct iwl_tso_hdr_page *hdr_page;
-	struct page **page_ptr;
 	struct tso_t tso;
 
 	/* if the packet is protected, then it must be CCMP or GCMP */
@@ -2130,14 +2158,11 @@ 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);
+	hdr_page = get_page_hdr(trans, hdr_room, skb);
 	if (!hdr_page)
 		return -ENOMEM;
 
-	get_page(hdr_page->page);
 	start_hdr = hdr_page->pos;
-	page_ptr = (void *)((u8 *)skb->cb + trans_pcie->page_offs);
-	*page_ptr = hdr_page->page;
 	memcpy(hdr_page->pos, skb->data + hdr_len, iv_len);
 	hdr_page->pos += iv_len;
 
@@ -2279,7 +2304,8 @@ static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
 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_cmd *dev_cmd, u16 tb1_len)
+				   struct iwl_device_tx_cmd *dev_cmd,
+				   u16 tb1_len)
 {
 	/* No A-MSDU without CONFIG_INET */
 	WARN_ON(1);
@@ -2289,7 +2315,7 @@ static int iwl_fill_data_tbs_amsdu(struct iwl_trans *trans, struct sk_buff *skb,
 #endif /* CONFIG_INET */
 
 int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
-		      struct iwl_device_cmd *dev_cmd, int txq_id)
+		      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;
@@ -2346,7 +2372,7 @@ int iwl_trans_pcie_tx(struct iwl_trans *trans, struct sk_buff *skb,
 
 		/* don't put the packet on the ring, if there is no room */
 		if (unlikely(iwl_queue_space(trans, txq) < 3)) {
-			struct iwl_device_cmd **dev_cmd_ptr;
+			struct iwl_device_tx_cmd **dev_cmd_ptr;
 
 			dev_cmd_ptr = (void *)((u8 *)skb->cb +
 					       trans_pcie->dev_cmd_offs);