iwlwifi: pcie: add AMSDU to gen2

This is essentially the same code as gen1, except that it uses
gen2 functions and SW checksum is not included.

Signed-off-by: Sara Sharon <sara.sharon@intel.com>
Signed-off-by: Luca Coelho <luciano.coelho@intel.com>

1 files changed, 168 insertions, 4 deletions

diff --git a/drivers/net/wireless/intel/iwlwifi/pcie/tx-gen2.c b/drivers/net/wireless/intel/iwlwifi/pcie/tx-gen2.c
index 9c9bfbbabdf1..6e186501f43c 100644
--- a/drivers/net/wireless/intel/iwlwifi/pcie/tx-gen2.c
+++ b/drivers/net/wireless/intel/iwlwifi/pcie/tx-gen2.c
@@ -49,6 +49,7 @@
  *
  *****************************************************************************/
 #include <linux/pm_runtime.h>
+#include <net/tso.h>
 
 #include "iwl-debug.h"
 #include "iwl-csr.h"
@@ -226,6 +227,143 @@ static int iwl_pcie_gen2_set_tb(struct iwl_trans *trans,
 	return idx;
 }
 
+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)
+{
+#ifdef CONFIG_INET
+	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;
+	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 */
+	iv_len = ieee80211_has_protected(hdr->frame_control) ?
+		IEEE80211_CCMP_HDR_LEN : 0;
+
+	trace_iwlwifi_dev_tx(trans->dev, skb, tfd, sizeof(*tfd),
+			     &dev_cmd->hdr, start_len, NULL, 0);
+
+	ip_hdrlen = skb_transport_header(skb) - skb_network_header(skb);
+	snap_ip_tcp_hdrlen = 8 + ip_hdrlen + tcp_hdrlen(skb);
+	total_len = skb->len - snap_ip_tcp_hdrlen - hdr_len - iv_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)) + iv_len;
+
+	/* Our device supports 9 segments at most, it will fit in 1 page */
+	hdr_page = get_page_hdr(trans, hdr_room);
+	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;
+
+	/*
+	 * Pull the ieee80211 header + IV to be able to use TSO core,
+	 * we will restore it for the tx_status flow.
+	 */
+	skb_pull(skb, hdr_len + iv_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);
+		struct sk_buff *csum_skb = NULL;
+		unsigned int tb_len;
+		dma_addr_t tb_phys;
+		struct tcphdr *tcph;
+		u8 *iph, *subf_hdrs_start = hdr_page->pos;
+
+		total_len -= data_left;
+
+		memset(hdr_page->pos, 0, amsdu_pad);
+		hdr_page->pos += 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;
+
+		length = snap_ip_tcp_hdrlen + data_left;
+		*((__be16 *)hdr_page->pos) = cpu_to_be16(length);
+		hdr_page->pos += 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);
+		iph = hdr_page->pos + 8;
+		tcph = (void *)(iph + ip_hdrlen);
+
+		hdr_page->pos += snap_ip_tcp_hdrlen;
+
+		tb_len = hdr_page->pos - start_hdr;
+		tb_phys = dma_map_single(trans->dev, start_hdr,
+					 tb_len, DMA_TO_DEVICE);
+		if (unlikely(dma_mapping_error(trans->dev, tb_phys))) {
+			dev_kfree_skb(csum_skb);
+			goto out_err;
+		}
+		iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb_len);
+		trace_iwlwifi_dev_tx_tso_chunk(trans->dev, start_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);
+
+		/* prepare the start_hdr for the next subframe */
+		start_hdr = hdr_page->pos;
+
+		/* put the payload */
+		while (data_left) {
+			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))) {
+				dev_kfree_skb(csum_skb);
+				goto out_err;
+			}
+			iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb_len);
+			trace_iwlwifi_dev_tx_tso_chunk(trans->dev, tso.data,
+						       tb_len);
+
+			data_left -= tb_len;
+			tso_build_data(skb, &tso, tb_len);
+		}
+	}
+
+	/* re -add the WiFi header and IV */
+	skb_push(skb, hdr_len + iv_len);
+
+	return 0;
+
+out_err:
+#endif
+	return -EINVAL;
+}
+
 static
 struct iwl_tfh_tfd *iwl_pcie_gen2_build_tfd(struct iwl_trans *trans,
 					    struct iwl_txq *txq,
@@ -238,15 +376,21 @@ struct iwl_tfh_tfd *iwl_pcie_gen2_build_tfd(struct iwl_trans *trans,
 	struct iwl_tfh_tfd *tfd =
 		iwl_pcie_get_tfd(trans_pcie, txq, txq->write_ptr);
 	dma_addr_t tb_phys;
+	bool amsdu;
 	int i, len, tb1_len, tb2_len, hdr_len;
 	void *tb1_addr;
 
 	memset(tfd, 0, sizeof(*tfd));
 
+	amsdu = ieee80211_is_data_qos(hdr->frame_control) &&
+		(*ieee80211_get_qos_ctl(hdr) &
+		 IEEE80211_QOS_CTL_A_MSDU_PRESENT);
+
 	tb_phys = iwl_pcie_get_first_tb_dma(txq, txq->write_ptr);
 	/* The first TB points to bi-directional DMA data */
-	memcpy(&txq->first_tb_bufs[txq->write_ptr], &dev_cmd->hdr,
-	       IWL_FIRST_TB_SIZE);
+	if (!amsdu)
+		memcpy(&txq->first_tb_bufs[txq->write_ptr], &dev_cmd->hdr,
+		       IWL_FIRST_TB_SIZE);
 
 	iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, IWL_FIRST_TB_SIZE);
 
@@ -262,7 +406,11 @@ struct iwl_tfh_tfd *iwl_pcie_gen2_build_tfd(struct iwl_trans *trans,
 	len = sizeof(struct iwl_tx_cmd_gen2) + sizeof(struct iwl_cmd_header) +
 	      ieee80211_hdrlen(hdr->frame_control) - IWL_FIRST_TB_SIZE;
 
-	tb1_len = ALIGN(len, 4);
+	/* do not align A-MSDU to dword as the subframe header aligns it */
+	if (amsdu)
+		tb1_len = len;
+	else
+		tb1_len = ALIGN(len, 4);
 
 	/* map the data for TB1 */
 	tb1_addr = ((u8 *)&dev_cmd->hdr) + IWL_FIRST_TB_SIZE;
@@ -271,8 +419,24 @@ struct iwl_tfh_tfd *iwl_pcie_gen2_build_tfd(struct iwl_trans *trans,
 		goto out_err;
 	iwl_pcie_gen2_set_tb(trans, tfd, tb_phys, tb1_len);
 
-	/* set up TFD's third entry to point to remainder of skb's head */
 	hdr_len = ieee80211_hdrlen(hdr->frame_control);
+
+	if (amsdu) {
+		if (!iwl_pcie_gen2_build_amsdu(trans, skb, tfd,
+					       tb1_len + IWL_FIRST_TB_SIZE,
+					       hdr_len, dev_cmd))
+			goto out_err;
+
+		/*
+		 * building the A-MSDU might have changed this data, so memcpy
+		 * it now
+		 */
+		memcpy(&txq->first_tb_bufs[txq->write_ptr], &dev_cmd->hdr,
+		       IWL_FIRST_TB_SIZE);
+		return tfd;
+	}
+
+	/* 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) {