iwlagn: support multiple TBs per command

The current "huge" command handling is a bit
confusing, and very limited since only one
command may be huge at a time. Additionally,
we often copy data around quite pointlessly
since we could instead map the existing scan
buffer for example and use it directly.

This patch makes that possible. The first
change is that multiple buffers may be given
to each command (this change was prepared
earlier so callsites don't need to change).
Each of those can be mapped attached to a TB
in the TFD, and the command header can use a
TB (the first one) in the TFD as well.

Doing this allows getting rid of huge commands
in favour of mapping existing buffers. The
beacon transmission is also optimised to not
copy the SKB at all but use multiple TBs.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: Wey-Yi Guy <wey-yi.w.guy@intel.com>

drivers/net/wireless/iwlwifi/iwl-agn-calib.c

@@ -87,7 +87,6 @@ int iwl_send_calib_results(struct iwl_priv *priv)
 
 	struct iwl_host_cmd hcmd = {
 		.id = REPLY_PHY_CALIBRATION_CMD,
-		.flags = CMD_SIZE_HUGE,
 	};
 
 	for (i = 0; i < IWL_CALIB_MAX; i++) {
@@ -95,6 +94,7 @@ int iwl_send_calib_results(struct iwl_priv *priv)
 		    priv->calib_results[i].buf) {
 			hcmd.len[0] = priv->calib_results[i].buf_len;
 			hcmd.data[0] = priv->calib_results[i].buf;
+			hcmd.dataflags[0] = IWL_HCMD_DFL_NOCOPY;
 			ret = iwl_send_cmd_sync(priv, &hcmd);
 			if (ret) {
 				IWL_ERR(priv, "Error %d iteration %d\n",

drivers/net/wireless/iwlwifi/iwl-agn-lib.c

@@ -1141,7 +1141,6 @@ int iwlagn_request_scan(struct iwl_priv *priv, struct ieee80211_vif *vif)
 	struct iwl_host_cmd cmd = {
 		.id = REPLY_SCAN_CMD,
 		.len = { sizeof(struct iwl_scan_cmd), },
-		.flags = CMD_SIZE_HUGE,
 	};
 	struct iwl_scan_cmd *scan;
 	struct iwl_rxon_context *ctx = &priv->contexts[IWL_RXON_CTX_BSS];
@@ -1428,6 +1427,7 @@ int iwlagn_request_scan(struct iwl_priv *priv, struct ieee80211_vif *vif)
 	cmd.len[0] += le16_to_cpu(scan->tx_cmd.len) +
 	    scan->channel_count * sizeof(struct iwl_scan_channel);
 	cmd.data[0] = scan;
+	cmd.dataflags[0] = IWL_HCMD_DFL_NOCOPY;
 	scan->len = cpu_to_le16(cmd.len[0]);
 
 	/* set scan bit here for PAN params */

drivers/net/wireless/iwlwifi/iwl-agn.c

@@ -134,12 +134,10 @@ int iwlagn_send_beacon_cmd(struct iwl_priv *priv)
 	struct iwl_tx_beacon_cmd *tx_beacon_cmd;
 	struct iwl_host_cmd cmd = {
 		.id = REPLY_TX_BEACON,
-		.flags = CMD_SIZE_HUGE,
 	};
 	u32 frame_size;
 	u32 rate_flags;
 	u32 rate;
-	int err;
 
 	/*
 	 * We have to set up the TX command, the TX Beacon command, and the
@@ -156,17 +154,15 @@ int iwlagn_send_beacon_cmd(struct iwl_priv *priv)
 	if (WARN_ON(!priv->beacon_skb))
 		return -EINVAL;
 
-	/* Allocate beacon memory */
-	tx_beacon_cmd = kzalloc(sizeof(*tx_beacon_cmd) + priv->beacon_skb->len,
-				GFP_KERNEL);
+	/* Allocate beacon command */
+	if (!priv->beacon_cmd)
+		priv->beacon_cmd = kzalloc(sizeof(*tx_beacon_cmd), GFP_KERNEL);
+	tx_beacon_cmd = priv->beacon_cmd;
 	if (!tx_beacon_cmd)
 		return -ENOMEM;
 
 	frame_size = priv->beacon_skb->len;
 
-	/* Set up TX beacon contents */
-	memcpy(tx_beacon_cmd->frame, priv->beacon_skb->data, frame_size);
-
 	/* Set up TX command fields */
 	tx_beacon_cmd->tx.len = cpu_to_le16((u16)frame_size);
 	tx_beacon_cmd->tx.sta_id = priv->beacon_ctx->bcast_sta_id;
@@ -175,7 +171,7 @@ int iwlagn_send_beacon_cmd(struct iwl_priv *priv)
 		TX_CMD_FLG_TSF_MSK | TX_CMD_FLG_STA_RATE_MSK;
 
 	/* Set up TX beacon command fields */
-	iwl_set_beacon_tim(priv, tx_beacon_cmd, (u8 *)tx_beacon_cmd->frame,
+	iwl_set_beacon_tim(priv, tx_beacon_cmd, priv->beacon_skb->data,
 			   frame_size);
 
 	/* Set up packet rate and flags */
@@ -189,15 +185,14 @@ int iwlagn_send_beacon_cmd(struct iwl_priv *priv)
 			rate_flags);
 
 	/* Submit command */
-	cmd.len[0] = sizeof(*tx_beacon_cmd) + frame_size;
+	cmd.len[0] = sizeof(*tx_beacon_cmd);
 	cmd.data[0] = tx_beacon_cmd;
+	cmd.dataflags[0] = IWL_HCMD_DFL_NOCOPY;
+	cmd.len[1] = frame_size;
+	cmd.data[1] = priv->beacon_skb->data;
+	cmd.dataflags[1] = IWL_HCMD_DFL_NOCOPY;
 
-	err = iwl_send_cmd_sync(priv, &cmd);
-
-	/* Free temporary storage */
-	kfree(tx_beacon_cmd);
-
-	return err;
+	return iwl_send_cmd_sync(priv, &cmd);
 }
 
 static void iwl_bg_beacon_update(struct work_struct *work)
@@ -3246,6 +3241,7 @@ static void iwl_uninit_drv(struct iwl_priv *priv)
 	iwlcore_free_geos(priv);
 	iwl_free_channel_map(priv);
 	kfree(priv->scan_cmd);
+	kfree(priv->beacon_cmd);
 }
 
 struct ieee80211_ops iwlagn_hw_ops = {

drivers/net/wireless/iwlwifi/iwl-commands.h

@@ -205,7 +205,6 @@ enum {
 #define QUEUE_TO_SEQ(q)	(((q) & 0x1f) << 8)
 #define SEQ_TO_INDEX(s)	((s) & 0xff)
 #define INDEX_TO_SEQ(i)	((i) & 0xff)
-#define SEQ_HUGE_FRAME	cpu_to_le16(0x4000)
 #define SEQ_RX_FRAME	cpu_to_le16(0x8000)
 
 /**
@@ -234,9 +233,7 @@ struct iwl_cmd_header {
 	 *
 	 *  0:7		tfd index - position within TX queue
 	 *  8:12	TX queue id
-	 *  13		reserved
-	 *  14		huge - driver sets this to indicate command is in the
-	 *  		'huge' storage at the end of the command buffers
+	 *  13:14	reserved
 	 *  15		unsolicited RX or uCode-originated notification
 	 */
 	__le16 sequence;

drivers/net/wireless/iwlwifi/iwl-dev.h

@@ -110,8 +110,6 @@ struct iwl_cmd_meta {
 			 struct iwl_device_cmd *cmd,
 			 struct iwl_rx_packet *pkt);
 
-	/* The CMD_SIZE_HUGE flag bit indicates that the command
-	 * structure is stored at the end of the shared queue memory. */
 	u32 flags;
 
 	DEFINE_DMA_UNMAP_ADDR(mapping);
@@ -121,7 +119,23 @@ struct iwl_cmd_meta {
 /*
  * Generic queue structure
  *
- * Contains common data for Rx and Tx queues
+ * Contains common data for Rx and Tx queues.
+ *
+ * Note the difference between n_bd and n_window: the hardware
+ * always assumes 256 descriptors, so n_bd is always 256 (unless
+ * there might be HW changes in the future). For the normal TX
+ * queues, n_window, which is the size of the software queue data
+ * is also 256; however, for the command queue, n_window is only
+ * 32 since we don't need so many commands pending. Since the HW
+ * still uses 256 BDs for DMA though, n_bd stays 256. As a result,
+ * the software buffers (in the variables @meta, @txb in struct
+ * iwl_tx_queue) only have 32 entries, while the HW buffers (@tfds
+ * in the same struct) have 256.
+ * This means that we end up with the following:
+ *  HW entries: | 0 | ... | N * 32 | ... | N * 32 + 31 | ... | 255 |
+ *  SW entries:           | 0      | ... | 31          |
+ * where N is a number between 0 and 7. This means that the SW
+ * data is a window overlayed over the HW queue.
  */
 struct iwl_queue {
 	int n_bd;              /* number of BDs in this queue */
@@ -163,7 +177,7 @@ struct iwl_tx_info {
 
 struct iwl_tx_queue {
 	struct iwl_queue q;
-	void *tfds;
+	struct iwl_tfd *tfds;
 	struct iwl_device_cmd **cmd;
 	struct iwl_cmd_meta *meta;
 	struct iwl_tx_info *txb;
@@ -245,7 +259,6 @@ enum {
 	CMD_SYNC = 0,
 	CMD_SIZE_NORMAL = 0,
 	CMD_NO_SKB = 0,
-	CMD_SIZE_HUGE = (1 << 0),
 	CMD_ASYNC = (1 << 1),
 	CMD_WANT_SKB = (1 << 2),
 	CMD_MAPPED = (1 << 3),
@@ -257,8 +270,8 @@ enum {
  * struct iwl_device_cmd
  *
  * For allocation of the command and tx queues, this establishes the overall
- * size of the largest command we send to uCode, except for a scan command
- * (which is relatively huge; space is allocated separately).
+ * size of the largest command we send to uCode, except for commands that
+ * aren't fully copied and use other TFD space.
  */
 struct iwl_device_cmd {
 	struct iwl_cmd_header hdr;	/* uCode API */
@@ -275,7 +288,11 @@ struct iwl_device_cmd {
 
 #define TFD_MAX_PAYLOAD_SIZE (sizeof(struct iwl_device_cmd))
 
-#define IWL_MAX_CMD_TFDS	1
+#define IWL_MAX_CMD_TFDS	2
+
+enum iwl_hcmd_dataflag {
+	IWL_HCMD_DFL_NOCOPY	= BIT(0),
+};
 
 struct iwl_host_cmd {
 	const void *data[IWL_MAX_CMD_TFDS];
@@ -285,6 +302,7 @@ struct iwl_host_cmd {
 			 struct iwl_rx_packet *pkt);
 	u32 flags;
 	u16 len[IWL_MAX_CMD_TFDS];
+	u8 dataflags[IWL_MAX_CMD_TFDS];
 	u8 id;
 };
 
@@ -687,17 +705,8 @@ static inline int iwl_queue_used(const struct iwl_queue *q, int i)
 }
 
 
-static inline u8 get_cmd_index(struct iwl_queue *q, u32 index, int is_huge)
+static inline u8 get_cmd_index(struct iwl_queue *q, u32 index)
 {
-	/*
-	 * This is for init calibration result and scan command which
-	 * required buffer > TFD_MAX_PAYLOAD_SIZE,
-	 * the big buffer at end of command array
-	 */
-	if (is_huge)
-		return q->n_window;	/* must be power of 2 */
-
-	/* Otherwise, use normal size buffers */
 	return index & (q->n_window - 1);
 }
 
@@ -1451,6 +1460,7 @@ struct iwl_priv {
 	struct work_struct beacon_update;
 	struct iwl_rxon_context *beacon_ctx;
 	struct sk_buff *beacon_skb;
+	void *beacon_cmd;
 
 	struct work_struct tt_work;
 	struct work_struct ct_enter;

drivers/net/wireless/iwlwifi/iwl-devtrace.h

@@ -137,20 +137,27 @@ TRACE_EVENT(iwlwifi_dev_ucode_wrap_event,
 #define TRACE_SYSTEM iwlwifi
 
 TRACE_EVENT(iwlwifi_dev_hcmd,
-	TP_PROTO(struct iwl_priv *priv, void *hcmd, size_t len, u32 flags),
-	TP_ARGS(priv, hcmd, len, flags),
+	TP_PROTO(struct iwl_priv *priv, u32 flags,
+		 const void *hcmd0, size_t len0,
+		 const void *hcmd1, size_t len1,
+		 const void *hcmd2, size_t len2),
+	TP_ARGS(priv, flags, hcmd0, len0, hcmd1, len1, hcmd2, len2),
 	TP_STRUCT__entry(
 		PRIV_ENTRY
-		__dynamic_array(u8, hcmd, len)
+		__dynamic_array(u8, hcmd0, len0)
+		__dynamic_array(u8, hcmd1, len1)
+		__dynamic_array(u8, hcmd2, len2)
 		__field(u32, flags)
 	),
 	TP_fast_assign(
 		PRIV_ASSIGN;
-		memcpy(__get_dynamic_array(hcmd), hcmd, len);
+		memcpy(__get_dynamic_array(hcmd0), hcmd0, len0);
+		memcpy(__get_dynamic_array(hcmd1), hcmd1, len1);
+		memcpy(__get_dynamic_array(hcmd2), hcmd2, len2);
 		__entry->flags = flags;
 	),
 	TP_printk("[%p] hcmd %#.2x (%ssync)",
-		  __entry->priv, ((u8 *)__get_dynamic_array(hcmd))[0],
+		  __entry->priv, ((u8 *)__get_dynamic_array(hcmd0))[0],
 		  __entry->flags & CMD_ASYNC ? "a" : "")
 );
 

drivers/net/wireless/iwlwifi/iwl-sv-open.c

@@ -200,6 +200,7 @@ static int iwl_testmode_ucode(struct ieee80211_hw *hw, struct nlattr **tb)
 	cmd.id = nla_get_u8(tb[IWL_TM_ATTR_UCODE_CMD_ID]);
 	cmd.data[0] = nla_data(tb[IWL_TM_ATTR_UCODE_CMD_DATA]);
 	cmd.len[0] = nla_len(tb[IWL_TM_ATTR_UCODE_CMD_DATA]);
+	cmd.dataflags[0] = IWL_HCMD_DFL_NOCOPY;
 	IWL_INFO(priv, "testmode ucode command ID 0x%x, flags 0x%x,"
 				" len %d\n", cmd.id, cmd.flags, cmd.len[0]);
 	/* ok, let's submit the command to ucode */