diff mbox

[v6,10/10] ath10k: sdio support

Message ID 148959281831.31432.17236410281378572164.stgit@potku.adurom.net (mailing list archive)
State Changes Requested
Delegated to: Kalle Valo
Headers show

Commit Message

Kalle Valo March 15, 2017, 3:46 p.m. UTC
From: Erik Stromdahl <erik.stromdahl@gmail.com>

sdio/mailbox HIF implementation.

Signed-off-by: Erik Stromdahl <erik.stromdahl@gmail.com>
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
---
 drivers/net/wireless/ath/ath10k/Kconfig  |    6 
 drivers/net/wireless/ath/ath10k/Makefile |    3 
 drivers/net/wireless/ath/ath10k/sdio.c   | 2102 ++++++++++++++++++++++++++++++
 drivers/net/wireless/ath/ath10k/sdio.h   |  263 ++++
 4 files changed, 2374 insertions(+)
 create mode 100644 drivers/net/wireless/ath/ath10k/sdio.c
 create mode 100644 drivers/net/wireless/ath/ath10k/sdio.h

Comments

Kalle Valo March 23, 2017, 3:23 p.m. UTC | #1
Kalle Valo <kvalo@qca.qualcomm.com> writes:

> From: Erik Stromdahl <erik.stromdahl@gmail.com>
>
> sdio/mailbox HIF implementation.
>
> Signed-off-by: Erik Stromdahl <erik.stromdahl@gmail.com>
> Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>

[...]

> +static int ath10k_sdio_mbox_proc_cpu_intr(struct ath10k *ar)
> +{
> +	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
> +	struct ath10k_sdio_irq_data *irq_data = &ar_sdio->irq_data;
> +	u8 cpu_int_status, *reg_buf;
> +	int ret;
> +
> +	mutex_lock(&irq_data->mtx);
> +	cpu_int_status = irq_data->irq_proc_reg->cpu_int_status &
> +			 irq_data->irq_en_reg->cpu_int_status_en;
> +	if (!cpu_int_status) {
> +		ath10k_warn(ar, "CPU interrupt status is zero\n");
> +		return -EIO;
> +	}
> +
> +	/* Clear the interrupt */
> +	irq_data->irq_proc_reg->cpu_int_status &= ~cpu_int_status;
> +
> +	/* Set up the register transfer buffer to hit the register 4 times ,
> +	 * this is done to make the access 4-byte aligned to mitigate issues
> +	 * with host bus interconnects that restrict bus transfer lengths to
> +	 * be a multiple of 4-bytes.
> +	 */
> +	reg_buf = kzalloc(4, GFP_KERNEL);
> +	if (!reg_buf)
> +		return -ENOMEM;

kbuild bot reported that I failed to release the mutex in these two
return cases above, I'll fix that in v7.
diff mbox

Patch

diff --git a/drivers/net/wireless/ath/ath10k/Kconfig b/drivers/net/wireless/ath/ath10k/Kconfig
index b4241cf9b7ed..f2f6321a1c00 100644
--- a/drivers/net/wireless/ath/ath10k/Kconfig
+++ b/drivers/net/wireless/ath/ath10k/Kconfig
@@ -22,6 +22,12 @@  config ATH10K_AHB
 	---help---
 	  This module adds support for AHB bus
 
+config ATH10K_SDIO
+	tristate "Atheros ath10k SDIO support (EXPERIMENTAL)"
+	depends on ATH10K && MMC
+	---help---
+	  This module adds support for SDIO/MMC bus
+
 config ATH10K_DEBUG
 	bool "Atheros ath10k debugging"
 	depends on ATH10K
diff --git a/drivers/net/wireless/ath/ath10k/Makefile b/drivers/net/wireless/ath/ath10k/Makefile
index 930fadd940d8..b0b19a7eb98b 100644
--- a/drivers/net/wireless/ath/ath10k/Makefile
+++ b/drivers/net/wireless/ath/ath10k/Makefile
@@ -27,5 +27,8 @@  ath10k_pci-y += pci.o \
 
 ath10k_pci-$(CONFIG_ATH10K_AHB) += ahb.o
 
+obj-$(CONFIG_ATH10K_SDIO) += ath10k_sdio.o
+ath10k_sdio-y += sdio.o
+
 # for tracing framework to find trace.h
 CFLAGS_trace.o := -I$(src)
diff --git a/drivers/net/wireless/ath/ath10k/sdio.c b/drivers/net/wireless/ath/ath10k/sdio.c
new file mode 100644
index 000000000000..bcb060a562bf
--- /dev/null
+++ b/drivers/net/wireless/ath/ath10k/sdio.c
@@ -0,0 +1,2102 @@ 
+/*
+ * Copyright (c) 2004-2011 Atheros Communications Inc.
+ * Copyright (c) 2011-2012 Qualcomm Atheros, Inc.
+ * Copyright (c) 2016-2017 Erik Stromdahl <erik.stromdahl@gmail.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <linux/mmc/card.h>
+#include <linux/mmc/mmc.h>
+#include <linux/mmc/host.h>
+#include <linux/mmc/sdio_func.h>
+#include <linux/mmc/sdio_ids.h>
+#include <linux/mmc/sdio.h>
+#include <linux/mmc/sd.h>
+#include <linux/bitfield.h>
+#include "core.h"
+#include "bmi.h"
+#include "debug.h"
+#include "hif.h"
+#include "htc.h"
+#include "targaddrs.h"
+#include "trace.h"
+#include "sdio.h"
+
+/* inlined helper functions */
+
+static inline int ath10k_sdio_calc_txrx_padded_len(struct ath10k_sdio *ar_sdio,
+						   size_t len)
+{
+	return __ALIGN_MASK((len), ar_sdio->mbox_info.block_mask);
+}
+
+static inline enum ath10k_htc_ep_id pipe_id_to_eid(u8 pipe_id)
+{
+	return (enum ath10k_htc_ep_id)pipe_id;
+}
+
+static inline void ath10k_sdio_mbox_free_rx_pkt(struct ath10k_sdio_rx_data *pkt)
+{
+	dev_kfree_skb(pkt->skb);
+	pkt->skb = NULL;
+	pkt->alloc_len = 0;
+	pkt->act_len = 0;
+	pkt->trailer_only = false;
+}
+
+static inline int ath10k_sdio_mbox_alloc_rx_pkt(struct ath10k_sdio_rx_data *pkt,
+						size_t act_len, size_t full_len,
+						bool part_of_bundle,
+						bool last_in_bundle)
+{
+	pkt->skb = dev_alloc_skb(full_len);
+	if (!pkt->skb)
+		return -ENOMEM;
+
+	pkt->act_len = act_len;
+	pkt->alloc_len = full_len;
+	pkt->part_of_bundle = part_of_bundle;
+	pkt->last_in_bundle = last_in_bundle;
+	pkt->trailer_only = false;
+
+	return 0;
+}
+
+static inline bool is_trailer_only_msg(struct ath10k_sdio_rx_data *pkt)
+{
+	bool trailer_only = false;
+	struct ath10k_htc_hdr *htc_hdr =
+		(struct ath10k_htc_hdr *)pkt->skb->data;
+	u16 len = __le16_to_cpu(htc_hdr->len);
+
+	if (len == htc_hdr->trailer_len)
+		trailer_only = true;
+
+	return trailer_only;
+}
+
+/* sdio/mmc functions */
+
+static inline void ath10k_sdio_set_cmd52_arg(u32 *arg, u8 write, u8 raw,
+					     unsigned int address,
+					     unsigned char val)
+{
+	*arg = FIELD_PREP(BIT(31), write) |
+	       FIELD_PREP(BIT(27), raw) |
+	       FIELD_PREP(BIT(26), 1) |
+	       FIELD_PREP(GENMASK(25, 9), address) |
+	       FIELD_PREP(BIT(8), 1) |
+	       FIELD_PREP(GENMASK(7, 0), val);
+}
+
+static int ath10k_sdio_func0_cmd52_wr_byte(struct mmc_card *card,
+					   unsigned int address,
+					   unsigned char byte)
+{
+	struct mmc_command io_cmd;
+
+	memset(&io_cmd, 0, sizeof(io_cmd));
+	ath10k_sdio_set_cmd52_arg(&io_cmd.arg, 1, 0, address, byte);
+	io_cmd.opcode = SD_IO_RW_DIRECT;
+	io_cmd.flags = MMC_RSP_R5 | MMC_CMD_AC;
+
+	return mmc_wait_for_cmd(card->host, &io_cmd, 0);
+}
+
+static int ath10k_sdio_func0_cmd52_rd_byte(struct mmc_card *card,
+					   unsigned int address,
+					   unsigned char *byte)
+{
+	struct mmc_command io_cmd;
+	int ret;
+
+	memset(&io_cmd, 0, sizeof(io_cmd));
+	ath10k_sdio_set_cmd52_arg(&io_cmd.arg, 0, 0, address, 0);
+	io_cmd.opcode = SD_IO_RW_DIRECT;
+	io_cmd.flags = MMC_RSP_R5 | MMC_CMD_AC;
+
+	ret = mmc_wait_for_cmd(card->host, &io_cmd, 0);
+	if (!ret)
+		*byte = io_cmd.resp[0];
+
+	return ret;
+}
+
+static int ath10k_sdio_config(struct ath10k *ar)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	struct sdio_func *func = ar_sdio->func;
+	unsigned char byte, asyncintdelay = 2;
+	int ret;
+
+	ath10k_dbg(ar, ATH10K_DBG_BOOT, "sdio configuration\n");
+
+	sdio_claim_host(func);
+
+	byte = 0;
+	ret = ath10k_sdio_func0_cmd52_rd_byte(func->card,
+					      SDIO_CCCR_DRIVE_STRENGTH,
+					      &byte);
+
+	byte &= ~ATH10K_SDIO_DRIVE_DTSX_MASK;
+	byte |= FIELD_PREP(ATH10K_SDIO_DRIVE_DTSX_MASK,
+			   ATH10K_SDIO_DRIVE_DTSX_TYPE_D);
+
+	ret = ath10k_sdio_func0_cmd52_wr_byte(func->card,
+					      SDIO_CCCR_DRIVE_STRENGTH,
+					      byte);
+
+	byte = 0;
+	ret = ath10k_sdio_func0_cmd52_rd_byte(
+		func->card,
+		CCCR_SDIO_DRIVER_STRENGTH_ENABLE_ADDR,
+		&byte);
+
+	byte |= (CCCR_SDIO_DRIVER_STRENGTH_ENABLE_A |
+		 CCCR_SDIO_DRIVER_STRENGTH_ENABLE_C |
+		 CCCR_SDIO_DRIVER_STRENGTH_ENABLE_D);
+
+	ret = ath10k_sdio_func0_cmd52_wr_byte(func->card,
+					      CCCR_SDIO_DRIVER_STRENGTH_ENABLE_ADDR,
+					      byte);
+	if (ret) {
+		ath10k_warn(ar, "failed to enable driver strength: %d\n", ret);
+		goto out;
+	}
+
+	byte = 0;
+	ret = ath10k_sdio_func0_cmd52_rd_byte(func->card,
+					      CCCR_SDIO_IRQ_MODE_REG_SDIO3,
+					      &byte);
+
+	byte |= SDIO_IRQ_MODE_ASYNC_4BIT_IRQ_SDIO3;
+
+	ret = ath10k_sdio_func0_cmd52_wr_byte(func->card,
+					      CCCR_SDIO_IRQ_MODE_REG_SDIO3,
+					      byte);
+	if (ret) {
+		ath10k_warn(ar, "failed to enable 4-bit async irq mode: %d\n",
+			    ret);
+		goto out;
+	}
+
+	byte = 0;
+	ret = ath10k_sdio_func0_cmd52_rd_byte(func->card,
+					      CCCR_SDIO_ASYNC_INT_DELAY_ADDRESS,
+					      &byte);
+
+	byte &= ~CCCR_SDIO_ASYNC_INT_DELAY_MASK;
+	byte |= FIELD_PREP(CCCR_SDIO_ASYNC_INT_DELAY_MASK, asyncintdelay);
+
+	ret = ath10k_sdio_func0_cmd52_wr_byte(func->card,
+					      CCCR_SDIO_ASYNC_INT_DELAY_ADDRESS,
+					      byte);
+
+	/* give us some time to enable, in ms */
+	func->enable_timeout = 100;
+
+	ret = sdio_set_block_size(func, ar_sdio->mbox_info.block_size);
+	if (ret) {
+		ath10k_warn(ar, "failed to set sdio block size to %d: %d\n",
+			    ar_sdio->mbox_info.block_size, ret);
+		goto out;
+	}
+
+out:
+	sdio_release_host(func);
+	return ret;
+}
+
+static int ath10k_sdio_io(struct ath10k *ar, u32 request, u32 addr,
+			  u8 *buf, u32 len)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	struct sdio_func *func = ar_sdio->func;
+	int ret;
+
+	sdio_claim_host(func);
+
+	if (request & HIF_WRITE) {
+		if (request & HIF_FIXED_ADDRESS)
+			ret = sdio_writesb(func, addr, buf, len);
+		else
+			ret = sdio_memcpy_toio(func, addr, buf, len);
+	} else {
+		if (request & HIF_FIXED_ADDRESS)
+			ret = sdio_readsb(func, buf, addr, len);
+		else
+			ret = sdio_memcpy_fromio(func, buf, addr, len);
+	}
+
+	sdio_release_host(func);
+
+	ath10k_dbg(ar, ATH10K_DBG_SDIO, "sdio %s addr 0x%x%s buf 0x%p len %d\n",
+		   request & HIF_WRITE ? "wr" : "rd", addr,
+		   request & HIF_FIXED_ADDRESS ? " (fixed)" : "", buf, len);
+	ath10k_dbg_dump(ar, ATH10K_DBG_SDIO_DUMP, NULL,
+			request & HIF_WRITE ? "sdio wr " : "sdio rd ",
+			buf, len);
+
+	return ret;
+}
+
+static int ath10k_sdio_read_write_sync(struct ath10k *ar, u32 addr, u8 *buf,
+				       u32 len, u32 request)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+
+	if (request & HIF_BLOCK_BASIS)
+		len = round_down(len, ar_sdio->mbox_info.block_size);
+
+	return ath10k_sdio_io(ar, request, addr, buf, len);
+}
+
+/* HIF mbox functions */
+
+static int ath10k_sdio_mbox_rx_process_packet(struct ath10k *ar,
+					      struct ath10k_sdio_rx_data *pkt,
+					      u32 *lookaheads,
+					      int *n_lookaheads)
+{
+	struct ath10k_htc *htc = &ar->htc;
+	struct sk_buff *skb = pkt->skb;
+	struct ath10k_htc_hdr *htc_hdr = (struct ath10k_htc_hdr *)skb->data;
+	bool trailer_present = htc_hdr->flags & ATH10K_HTC_FLAG_TRAILER_PRESENT;
+	enum ath10k_htc_ep_id eid;
+	u16 payload_len;
+	u8 *trailer;
+	int ret;
+
+	payload_len = le16_to_cpu(htc_hdr->len);
+
+	if (trailer_present) {
+		trailer = skb->data + sizeof(*htc_hdr) +
+			  payload_len - htc_hdr->trailer_len;
+
+		eid = pipe_id_to_eid(htc_hdr->eid);
+
+		ret = ath10k_htc_process_trailer(htc,
+						 trailer,
+						 htc_hdr->trailer_len,
+						 eid,
+						 lookaheads,
+						 n_lookaheads);
+		if (ret)
+			return ret;
+
+		if (is_trailer_only_msg(pkt))
+			pkt->trailer_only = true;
+
+		skb_trim(skb, skb->len - htc_hdr->trailer_len);
+	}
+
+	skb_pull(skb, sizeof(*htc_hdr));
+
+	return 0;
+}
+
+static int ath10k_sdio_mbox_rx_process_packets(struct ath10k *ar,
+					       u32 lookaheads[],
+					       int *n_lookahead)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	struct ath10k_htc *htc = &ar->htc;
+	struct ath10k_sdio_rx_data *pkt;
+	struct ath10k_htc_ep *ep;
+	enum ath10k_htc_ep_id id;
+	int ret, i, *n_lookahead_local;
+	u32 *lookaheads_local;
+
+	for (i = 0; i < ar_sdio->n_rx_pkts; i++) {
+		lookaheads_local = lookaheads;
+		n_lookahead_local = n_lookahead;
+
+		id = ((struct ath10k_htc_hdr *)&lookaheads[i])->eid;
+
+		if (id >= ATH10K_HTC_EP_COUNT) {
+			ath10k_warn(ar, "invalid endpoint in look-ahead: %d\n",
+				    id);
+			ret = -ENOMEM;
+			goto out;
+		}
+
+		ep = &htc->endpoint[id];
+
+		if (ep->service_id == 0) {
+			ath10k_warn(ar, "ep %d is not connected\n", id);
+			ret = -ENOMEM;
+			goto out;
+		}
+
+		pkt = &ar_sdio->rx_pkts[i];
+
+		if (pkt->part_of_bundle && !pkt->last_in_bundle) {
+			/* Only read lookahead's from RX trailers
+			 * for the last packet in a bundle.
+			 */
+			lookaheads_local = NULL;
+			n_lookahead_local = NULL;
+		}
+
+		ret = ath10k_sdio_mbox_rx_process_packet(ar,
+							 pkt,
+							 lookaheads_local,
+							 n_lookahead_local);
+		if (ret)
+			goto out;
+
+		if (!pkt->trailer_only)
+			ep->ep_ops.ep_rx_complete(ar_sdio->ar, pkt->skb);
+		else
+			kfree_skb(pkt->skb);
+
+		/* The RX complete handler now owns the skb...*/
+		pkt->skb = NULL;
+		pkt->alloc_len = 0;
+	}
+
+	ret = 0;
+
+out:
+	/* Free all packets that was not passed on to the RX completion
+	 * handler...
+	 */
+	for (; i < ar_sdio->n_rx_pkts; i++)
+		ath10k_sdio_mbox_free_rx_pkt(&ar_sdio->rx_pkts[i]);
+
+	return ret;
+}
+
+static int ath10k_sdio_mbox_alloc_pkt_bundle(struct ath10k *ar,
+					     struct ath10k_sdio_rx_data *rx_pkts,
+					     struct ath10k_htc_hdr *htc_hdr,
+					     size_t full_len, size_t act_len,
+					     size_t *bndl_cnt)
+{
+	int ret, i;
+
+	*bndl_cnt = FIELD_GET(ATH10K_HTC_FLAG_BUNDLE_MASK, htc_hdr->flags);
+
+	if (*bndl_cnt > HTC_HOST_MAX_MSG_PER_BUNDLE) {
+		ath10k_warn(ar,
+			    "HTC bundle length %u exceeds maximum %u\n",
+			    le16_to_cpu(htc_hdr->len),
+			    HTC_HOST_MAX_MSG_PER_BUNDLE);
+		return -ENOMEM;
+	}
+
+	/* Allocate bndl_cnt extra skb's for the bundle.
+	 * The package containing the
+	 * ATH10K_HTC_FLAG_BUNDLE_MASK flag is not included
+	 * in bndl_cnt. The skb for that packet will be
+	 * allocated separately.
+	 */
+	for (i = 0; i < *bndl_cnt; i++) {
+		ret = ath10k_sdio_mbox_alloc_rx_pkt(&rx_pkts[i],
+						    act_len,
+						    full_len,
+						    true,
+						    false);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+static int ath10k_sdio_mbox_rx_alloc(struct ath10k *ar,
+				     u32 lookaheads[], int n_lookaheads)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	struct ath10k_htc_hdr *htc_hdr;
+	size_t full_len, act_len;
+	bool last_in_bundle;
+	int ret, i;
+
+	if (n_lookaheads > ATH10K_SDIO_MAX_RX_MSGS) {
+		ath10k_warn(ar,
+			    "the total number of pkgs to be fetched (%u) exceeds maximum %u\n",
+			    n_lookaheads,
+			    ATH10K_SDIO_MAX_RX_MSGS);
+		ret = -ENOMEM;
+		goto err;
+	}
+
+	for (i = 0; i < n_lookaheads; i++) {
+		htc_hdr = (struct ath10k_htc_hdr *)&lookaheads[i];
+		last_in_bundle = false;
+
+		if (le16_to_cpu(htc_hdr->len) >
+		    ATH10K_HTC_MBOX_MAX_PAYLOAD_LENGTH) {
+			ath10k_warn(ar,
+				    "payload length %d exceeds max htc length: %zu\n",
+				    le16_to_cpu(htc_hdr->len),
+				    ATH10K_HTC_MBOX_MAX_PAYLOAD_LENGTH);
+			ret = -ENOMEM;
+			goto err;
+		}
+
+		act_len = le16_to_cpu(htc_hdr->len) + sizeof(*htc_hdr);
+		full_len = ath10k_sdio_calc_txrx_padded_len(ar_sdio, act_len);
+
+		if (full_len > ATH10K_SDIO_MAX_BUFFER_SIZE) {
+			ath10k_warn(ar,
+				    "rx buffer requested with invalid htc_hdr length (%d, 0x%x): %d\n",
+				    htc_hdr->eid, htc_hdr->flags,
+				    le16_to_cpu(htc_hdr->len));
+			ret = -EINVAL;
+			goto err;
+		}
+
+		if (htc_hdr->flags & ATH10K_HTC_FLAG_BUNDLE_MASK) {
+			/* HTC header indicates that every packet to follow
+			 * has the same padded length so that it can be
+			 * optimally fetched as a full bundle.
+			 */
+			size_t bndl_cnt;
+
+			ret = ath10k_sdio_mbox_alloc_pkt_bundle(ar,
+								&ar_sdio->rx_pkts[i],
+								htc_hdr,
+								full_len,
+								act_len,
+								&bndl_cnt);
+
+			n_lookaheads += bndl_cnt;
+			i += bndl_cnt;
+			/*Next buffer will be the last in the bundle */
+			last_in_bundle = true;
+		}
+
+		/* Allocate skb for packet. If the packet had the
+		 * ATH10K_HTC_FLAG_BUNDLE_MASK flag set, all bundled
+		 * packet skb's have been allocated in the previous step.
+		 */
+		ret = ath10k_sdio_mbox_alloc_rx_pkt(&ar_sdio->rx_pkts[i],
+						    act_len,
+						    full_len,
+						    last_in_bundle,
+						    last_in_bundle);
+	}
+
+	ar_sdio->n_rx_pkts = i;
+
+	return 0;
+
+err:
+	for (i = 0; i < ATH10K_SDIO_MAX_RX_MSGS; i++) {
+		if (!ar_sdio->rx_pkts[i].alloc_len)
+			break;
+		ath10k_sdio_mbox_free_rx_pkt(&ar_sdio->rx_pkts[i]);
+	}
+
+	return ret;
+}
+
+static int ath10k_sdio_mbox_rx_packet(struct ath10k *ar,
+				      struct ath10k_sdio_rx_data *pkt)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	struct sk_buff *skb = pkt->skb;
+	int ret;
+
+	ret = ath10k_sdio_read_write_sync(ar,
+					  ar_sdio->mbox_info.htc_addr,
+					  skb->data, pkt->alloc_len,
+					  HIF_RD_SYNC_BLOCK_FIX);
+
+	pkt->status = ret;
+	if (!ret)
+		skb_put(skb, pkt->act_len);
+
+	return ret;
+}
+
+static int ath10k_sdio_mbox_rx_fetch(struct ath10k *ar)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	int ret, i;
+
+	for (i = 0; i < ar_sdio->n_rx_pkts; i++) {
+		ret = ath10k_sdio_mbox_rx_packet(ar,
+						 &ar_sdio->rx_pkts[i]);
+		if (ret)
+			goto err;
+	}
+
+	return 0;
+
+err:
+	/* Free all packets that was not successfully fetched. */
+	for (; i < ar_sdio->n_rx_pkts; i++)
+		ath10k_sdio_mbox_free_rx_pkt(&ar_sdio->rx_pkts[i]);
+
+	return ret;
+}
+
+/* This is the timeout for mailbox processing done in the sdio irq
+ * handler. The timeout is deliberately set quite high since SDIO dump logs
+ * over serial port can/will add a substantial overhead to the processing
+ * (if enabled).
+ */
+#define SDIO_MBOX_PROCESSING_TIMEOUT_HZ (20 * HZ)
+
+static int ath10k_sdio_mbox_rxmsg_pending_handler(struct ath10k *ar,
+						  u32 msg_lookahead, bool *done)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	u32 lookaheads[ATH10K_SDIO_MAX_RX_MSGS];
+	int n_lookaheads = 1;
+	unsigned long timeout;
+	int ret;
+
+	*done = true;
+
+	/* Copy the lookahead obtained from the HTC register table into our
+	 * temp array as a start value.
+	 */
+	lookaheads[0] = msg_lookahead;
+
+	timeout = jiffies + SDIO_MBOX_PROCESSING_TIMEOUT_HZ;
+	while (time_before(jiffies, timeout)) {
+		/* Try to allocate as many HTC RX packets indicated by
+		 * n_lookaheads.
+		 */
+		ret = ath10k_sdio_mbox_rx_alloc(ar, lookaheads,
+						n_lookaheads);
+		if (ret)
+			break;
+
+		if (ar_sdio->n_rx_pkts >= 2)
+			/* A recv bundle was detected, force IRQ status
+			 * re-check again.
+			 */
+			*done = false;
+
+		ret = ath10k_sdio_mbox_rx_fetch(ar);
+
+		/* Process fetched packets. This will potentially update
+		 * n_lookaheads depending on if the packets contain lookahead
+		 * reports.
+		 */
+		n_lookaheads = 0;
+		ret = ath10k_sdio_mbox_rx_process_packets(ar,
+							  lookaheads,
+							  &n_lookaheads);
+
+		if (!n_lookaheads || ret)
+			break;
+
+		/* For SYNCH processing, if we get here, we are running
+		 * through the loop again due to updated lookaheads. Set
+		 * flag that we should re-check IRQ status registers again
+		 * before leaving IRQ processing, this can net better
+		 * performance in high throughput situations.
+		 */
+		*done = false;
+	}
+
+	if (ret && (ret != -ECANCELED))
+		ath10k_warn(ar, "failed to get pending recv messages: %d\n",
+			    ret);
+
+	return ret;
+}
+
+static int ath10k_sdio_mbox_proc_dbg_intr(struct ath10k *ar)
+{
+	u32 *dummy;
+	int ret;
+
+	dummy = kzalloc(sizeof(*dummy), GFP_KERNEL);
+	if (!dummy)
+		return -ENOMEM;
+
+	/* TODO: Add firmware crash handling */
+	ath10k_warn(ar, "firmware crashed\n");
+
+	/* read counter to clear the interrupt, the debug error interrupt is
+	 * counter 0.
+	 */
+	ret = ath10k_sdio_read_write_sync(ar, MBOX_COUNT_DEC_ADDRESS,
+					  (u8 *)dummy, sizeof(*dummy),
+					  HIF_RD_SYNC_BYTE_INC);
+	if (ret)
+		ath10k_warn(ar, "failed to clear debug interrupt: %d\n", ret);
+
+	kfree(dummy);
+
+	return ret;
+}
+
+static int ath10k_sdio_mbox_proc_counter_intr(struct ath10k *ar)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	struct ath10k_sdio_irq_data *irq_data = &ar_sdio->irq_data;
+	u8 counter_int_status;
+	int ret;
+
+	mutex_lock(&irq_data->mtx);
+	counter_int_status = irq_data->irq_proc_reg->counter_int_status &
+			     irq_data->irq_en_reg->cntr_int_status_en;
+
+	/* NOTE: other modules like GMBOX may use the counter interrupt for
+	 * credit flow control on other counters, we only need to check for
+	 * the debug assertion counter interrupt.
+	 */
+	if (counter_int_status & ATH10K_SDIO_TARGET_DEBUG_INTR_MASK)
+		ret = ath10k_sdio_mbox_proc_dbg_intr(ar);
+	else
+		ret = 0;
+
+	mutex_unlock(&irq_data->mtx);
+
+	return ret;
+}
+
+static int ath10k_sdio_mbox_proc_err_intr(struct ath10k *ar)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	struct ath10k_sdio_irq_data *irq_data = &ar_sdio->irq_data;
+	u8 error_int_status, *reg_buf;
+	int ret;
+
+	reg_buf = kzalloc(4, GFP_KERNEL);
+	if (!reg_buf)
+		return -ENOMEM;
+
+	ath10k_dbg(ar, ATH10K_DBG_SDIO, "sdio error interrupt\n");
+
+	error_int_status = irq_data->irq_proc_reg->error_int_status & 0x0F;
+	if (!error_int_status) {
+		ath10k_warn(ar, "invalid error interrupt status: 0x%x\n",
+			    error_int_status);
+		ret = -EIO;
+		goto out;
+	}
+
+	ath10k_dbg(ar, ATH10K_DBG_SDIO,
+		   "sdio error_int_status 0x%x\n", error_int_status);
+
+	if (FIELD_GET(MBOX_ERROR_INT_STATUS_WAKEUP_MASK,
+		      error_int_status))
+		ath10k_dbg(ar, ATH10K_DBG_SDIO, "sdio interrupt error wakeup\n");
+
+	if (FIELD_GET(MBOX_ERROR_INT_STATUS_RX_UNDERFLOW_MASK,
+		      error_int_status))
+		ath10k_warn(ar, "rx underflow interrupt error\n");
+
+	if (FIELD_GET(MBOX_ERROR_INT_STATUS_TX_OVERFLOW_MASK,
+		      error_int_status))
+		ath10k_warn(ar, "tx overflow interrupt error\n");
+
+	/* Clear the interrupt */
+	irq_data->irq_proc_reg->error_int_status &= ~error_int_status;
+
+	/* set W1C value to clear the interrupt, this hits the register first */
+	reg_buf[0] = error_int_status;
+
+	ret = ath10k_sdio_read_write_sync(ar,
+					  MBOX_ERROR_INT_STATUS_ADDRESS,
+					  reg_buf, 4, HIF_WR_SYNC_BYTE_FIX);
+	if (ret) {
+		ath10k_warn(ar, "unable to write to error int status address: %d\n",
+			    ret);
+		goto out;
+	}
+
+out:
+	kfree(reg_buf);
+
+	return ret;
+}
+
+static int ath10k_sdio_mbox_proc_cpu_intr(struct ath10k *ar)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	struct ath10k_sdio_irq_data *irq_data = &ar_sdio->irq_data;
+	u8 cpu_int_status, *reg_buf;
+	int ret;
+
+	mutex_lock(&irq_data->mtx);
+	cpu_int_status = irq_data->irq_proc_reg->cpu_int_status &
+			 irq_data->irq_en_reg->cpu_int_status_en;
+	if (!cpu_int_status) {
+		ath10k_warn(ar, "CPU interrupt status is zero\n");
+		return -EIO;
+	}
+
+	/* Clear the interrupt */
+	irq_data->irq_proc_reg->cpu_int_status &= ~cpu_int_status;
+
+	/* Set up the register transfer buffer to hit the register 4 times ,
+	 * this is done to make the access 4-byte aligned to mitigate issues
+	 * with host bus interconnects that restrict bus transfer lengths to
+	 * be a multiple of 4-bytes.
+	 */
+	reg_buf = kzalloc(4, GFP_KERNEL);
+	if (!reg_buf)
+		return -ENOMEM;
+
+	/* set W1C value to clear the interrupt, this hits the register first */
+	reg_buf[0] = cpu_int_status;
+
+	ret = ath10k_sdio_read_write_sync(ar,
+					  MBOX_CPU_INT_STATUS_ADDRESS,
+					  reg_buf, 4, HIF_WR_SYNC_BYTE_FIX);
+	if (ret) {
+		ath10k_warn(ar, "unable to write to cpu interrupt status address: %d\n",
+			    ret);
+		goto out;
+	}
+
+out:
+	kfree(reg_buf);
+
+	mutex_unlock(&irq_data->mtx);
+	return ret;
+}
+
+static int ath10k_sdio_mbox_read_int_status(struct ath10k *ar,
+					    u8 *host_int_status,
+					    u32 *lookahead)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	struct ath10k_sdio_irq_data *irq_data = &ar_sdio->irq_data;
+	struct ath10k_sdio_irq_proc_regs *irq_proc_reg = irq_data->irq_proc_reg;
+	struct ath10k_sdio_irq_enable_regs *irq_en_reg = irq_data->irq_en_reg;
+	u8 htc_mbox = FIELD_PREP(ATH10K_HTC_MAILBOX_MASK, 1);
+	int ret;
+
+	mutex_lock(&irq_data->mtx);
+
+	*lookahead = 0;
+	*host_int_status = 0;
+
+	/* int_status_en is supposed to be non zero, otherwise interrupts
+	 * shouldn't be enabled. There is however a short time frame during
+	 * initialization between the irq register and int_status_en init
+	 * where this can happen.
+	 * We silently ignore this condition.
+	 */
+	if (!irq_en_reg->int_status_en) {
+		ret = 0;
+		goto out;
+	}
+
+	/* Read the first sizeof(struct ath10k_irq_proc_registers)
+	 * bytes of the HTC register table. This
+	 * will yield us the value of different int status
+	 * registers and the lookahead registers.
+	 */
+	ret = ath10k_sdio_read_write_sync(ar,
+					  MBOX_HOST_INT_STATUS_ADDRESS,
+					  (u8 *)irq_proc_reg,
+					  sizeof(*irq_proc_reg),
+					  HIF_RD_SYNC_BYTE_INC);
+	if (ret)
+		goto out;
+
+	/* Update only those registers that are enabled */
+	*host_int_status = irq_proc_reg->host_int_status &
+			   irq_en_reg->int_status_en;
+
+	/* Look at mbox status */
+	if (!(*host_int_status & htc_mbox)) {
+		*lookahead = 0;
+		/* FIXME: ret = 0 ? it's already set to that, but better be
+		 * explicit*/
+		goto out;
+	}
+
+	/* Mask out pending mbox value, we use look ahead as
+	 * the real flag for mbox processing.
+	 */
+	*host_int_status &= ~htc_mbox;
+	if (irq_proc_reg->rx_lookahead_valid & htc_mbox) {
+		*lookahead = le32_to_cpu(
+			irq_proc_reg->rx_lookahead[ATH10K_HTC_MAILBOX]);
+		if (!*lookahead)
+			ath10k_warn(ar, "sdio mbox lookahead is zero\n");
+	}
+
+out:
+	mutex_unlock(&irq_data->mtx);
+	return ret;
+}
+
+static int ath10k_sdio_mbox_proc_pending_irqs(struct ath10k *ar,
+					      bool *done)
+{
+	u8 host_int_status;
+	u32 lookahead;
+	int ret;
+
+	/* NOTE: HIF implementation guarantees that the context of this
+	 * call allows us to perform SYNCHRONOUS I/O, that is we can block,
+	 * sleep or call any API that can block or switch thread/task
+	 * contexts. This is a fully schedulable context.
+	 */
+
+	ret = ath10k_sdio_mbox_read_int_status(ar,
+					       &host_int_status,
+					       &lookahead);
+	if (ret) {
+		*done = true;
+		goto out;
+	}
+
+	if (!host_int_status && !lookahead) {
+		ret = 0;
+		*done = true;
+		goto out;
+	}
+
+	if (lookahead) {
+		ath10k_dbg(ar, ATH10K_DBG_SDIO,
+			   "sdio pending mailbox msg lookahead 0x%08x\n",
+			   lookahead);
+
+		ret = ath10k_sdio_mbox_rxmsg_pending_handler(ar,
+							     lookahead,
+							     done);
+		if (ret)
+			goto out;
+	}
+
+	/* now, handle the rest of the interrupts */
+	ath10k_dbg(ar, ATH10K_DBG_SDIO,
+		   "sdio host_int_status 0x%x\n", host_int_status);
+
+	if (FIELD_GET(MBOX_HOST_INT_STATUS_CPU_MASK, host_int_status)) {
+		/* CPU Interrupt */
+		ret = ath10k_sdio_mbox_proc_cpu_intr(ar);
+		if (ret)
+			goto out;
+	}
+
+	if (FIELD_GET(MBOX_HOST_INT_STATUS_ERROR_MASK, host_int_status)) {
+		/* Error Interrupt */
+		ret = ath10k_sdio_mbox_proc_err_intr(ar);
+		if (ret)
+			goto out;
+	}
+
+	if (FIELD_GET(MBOX_HOST_INT_STATUS_COUNTER_MASK, host_int_status))
+		/* Counter Interrupt */
+		ret = ath10k_sdio_mbox_proc_counter_intr(ar);
+
+	ret = 0;
+
+out:
+	/* An optimization to bypass reading the IRQ status registers
+	 * unecessarily which can re-wake the target, if upper layers
+	 * determine that we are in a low-throughput mode, we can rely on
+	 * taking another interrupt rather than re-checking the status
+	 * registers which can re-wake the target.
+	 *
+	 * NOTE : for host interfaces that makes use of detecting pending
+	 * mbox messages at hif can not use this optimization due to
+	 * possible side effects, SPI requires the host to drain all
+	 * messages from the mailbox before exiting the ISR routine.
+	 */
+
+	ath10k_dbg(ar, ATH10K_DBG_SDIO,
+		   "sdio pending irqs done %d status %d",
+		   *done, ret);
+
+	return ret;
+}
+
+static void ath10k_sdio_set_mbox_info(struct ath10k *ar)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	struct ath10k_mbox_info *mbox_info = &ar_sdio->mbox_info;
+	u16 device = ar_sdio->func->device, dev_id_base, dev_id_chiprev;
+
+	mbox_info->htc_addr = ATH10K_HIF_MBOX_BASE_ADDR;
+	mbox_info->block_size = ATH10K_HIF_MBOX_BLOCK_SIZE;
+	mbox_info->block_mask = ATH10K_HIF_MBOX_BLOCK_SIZE - 1;
+	mbox_info->gmbox_addr = ATH10K_HIF_GMBOX_BASE_ADDR;
+	mbox_info->gmbox_sz = ATH10K_HIF_GMBOX_WIDTH;
+
+	mbox_info->ext_info[0].htc_ext_addr = ATH10K_HIF_MBOX0_EXT_BASE_ADDR;
+
+	dev_id_base = FIELD_GET(QCA_MANUFACTURER_ID_BASE, device);
+	dev_id_chiprev = FIELD_GET(QCA_MANUFACTURER_ID_REV_MASK, device);
+	switch (dev_id_base) {
+	case QCA_MANUFACTURER_ID_AR6005_BASE:
+		if (dev_id_chiprev < 4)
+			mbox_info->ext_info[0].htc_ext_sz =
+				ATH10K_HIF_MBOX0_EXT_WIDTH;
+		else
+			/* from QCA6174 2.0(0x504), the width has been extended
+			 * to 56K
+			 */
+			mbox_info->ext_info[0].htc_ext_sz =
+				ATH10K_HIF_MBOX0_EXT_WIDTH_ROME_2_0;
+		break;
+	case QCA_MANUFACTURER_ID_QCA9377_BASE:
+		mbox_info->ext_info[0].htc_ext_sz =
+			ATH10K_HIF_MBOX0_EXT_WIDTH_ROME_2_0;
+		break;
+	default:
+		mbox_info->ext_info[0].htc_ext_sz =
+				ATH10K_HIF_MBOX0_EXT_WIDTH;
+	}
+
+	mbox_info->ext_info[1].htc_ext_addr =
+		mbox_info->ext_info[0].htc_ext_addr +
+		mbox_info->ext_info[0].htc_ext_sz +
+		ATH10K_HIF_MBOX_DUMMY_SPACE_SIZE;
+	mbox_info->ext_info[1].htc_ext_sz = ATH10K_HIF_MBOX1_EXT_WIDTH;
+}
+
+/* BMI functions */
+
+static int ath10k_sdio_bmi_credits(struct ath10k *ar)
+{
+	int ret;
+	u32 addr, *cmd_credits;
+	unsigned long timeout;
+
+	cmd_credits = kzalloc(sizeof(*cmd_credits), GFP_KERNEL);
+	if (!cmd_credits)
+		return -ENOMEM;
+
+	/* Read the counter register to get the command credits */
+	addr = MBOX_COUNT_DEC_ADDRESS + ATH10K_HIF_MBOX_NUM_MAX * 4;
+
+	timeout = jiffies + BMI_COMMUNICATION_TIMEOUT_HZ;
+	while (time_before(jiffies, timeout) && !*cmd_credits) {
+		/* Hit the credit counter with a 4-byte access, the first byte
+		 * read will hit the counter and cause a decrement, while the
+		 * remaining 3 bytes has no effect. The rationale behind this
+		 * is to make all HIF accesses 4-byte aligned.
+		 */
+		ret = ath10k_sdio_read_write_sync(ar, addr,
+						  (u8 *)cmd_credits,
+						  sizeof(*cmd_credits),
+						  HIF_RD_SYNC_BYTE_INC);
+		if (ret) {
+			ath10k_warn(ar,
+				    "unable to decrement the command credit count register: %d\n",
+				    ret);
+			goto out;
+		}
+
+		/* The counter is only 8 bits.
+		 * Ignore anything in the upper 3 bytes
+		 */
+		*cmd_credits &= 0xFF;
+	}
+
+	if (!*cmd_credits) {
+		ath10k_warn(ar, "bmi communication timeout\n");
+		ret = -ETIMEDOUT;
+		goto out;
+	}
+
+	ret = 0;
+
+out:
+	kfree(cmd_credits);
+
+	return ret;
+}
+
+static int ath10k_sdio_bmi_get_rx_lookahead(struct ath10k *ar)
+{
+	unsigned long timeout;
+	u32 *rx_word;
+	int ret;
+
+	rx_word = kzalloc(sizeof(*rx_word), GFP_KERNEL);
+	if (!rx_word)
+		return -ENOMEM;
+
+	timeout = jiffies + BMI_COMMUNICATION_TIMEOUT_HZ;
+	while ((time_before(jiffies, timeout)) && !*rx_word) {
+		ret = ath10k_sdio_read_write_sync(ar,
+						  MBOX_HOST_INT_STATUS_ADDRESS,
+						  (u8 *)rx_word,
+						  sizeof(*rx_word),
+						  HIF_RD_SYNC_BYTE_INC);
+		if (ret) {
+			ath10k_warn(ar, "unable to read RX_LOOKAHEAD_VALID: %d\n", ret);
+			goto out;
+		}
+
+		 /* all we really want is one bit */
+		*rx_word &= 1;
+	}
+
+	if (!*rx_word) {
+		ath10k_warn(ar, "bmi_recv_buf FIFO empty\n");
+		ret = -EINVAL;
+		goto out;
+	}
+
+out:
+	kfree(rx_word);
+
+	return ret;
+}
+
+static int ath10k_sdio_bmi_exchange_msg(struct ath10k *ar,
+					void *req, u32 req_len,
+					void *resp, u32 *resp_len)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	u32 addr;
+	int ret;
+
+	if (req) {
+		ret = ath10k_sdio_bmi_credits(ar);
+		if (ret)
+			return ret;
+
+		addr = ar_sdio->mbox_info.htc_addr;
+
+		memcpy(ar_sdio->bmi_buf, req, req_len);
+		ret = ath10k_sdio_read_write_sync(ar, addr, ar_sdio->bmi_buf,
+						  req_len,
+						  HIF_WR_SYNC_BYTE_INC);
+		if (ret) {
+			ath10k_warn(ar,
+				    "unable to send the bmi data to the device: %d\n",
+				    ret);
+			return ret;
+		}
+	}
+
+	if (!resp || !resp_len)
+		/* No response expected */
+		return 0;
+
+	/* During normal bootup, small reads may be required.
+	 * Rather than issue an HIF Read and then wait as the Target
+	 * adds successive bytes to the FIFO, we wait here until
+	 * we know that response data is available.
+	 *
+	 * This allows us to cleanly timeout on an unexpected
+	 * Target failure rather than risk problems at the HIF level.
+	 * In particular, this avoids SDIO timeouts and possibly garbage
+	 * data on some host controllers.  And on an interconnect
+	 * such as Compact Flash (as well as some SDIO masters) which
+	 * does not provide any indication on data timeout, it avoids
+	 * a potential hang or garbage response.
+	 *
+	 * Synchronization is more difficult for reads larger than the
+	 * size of the MBOX FIFO (128B), because the Target is unable
+	 * to push the 129th byte of data until AFTER the Host posts an
+	 * HIF Read and removes some FIFO data.  So for large reads the
+	 * Host proceeds to post an HIF Read BEFORE all the data is
+	 * actually available to read.  Fortunately, large BMI reads do
+	 * not occur in practice -- they're supported for debug/development.
+	 *
+	 * So Host/Target BMI synchronization is divided into these cases:
+	 *  CASE 1: length < 4
+	 *        Should not happen
+	 *
+	 *  CASE 2: 4 <= length <= 128
+	 *        Wait for first 4 bytes to be in FIFO
+	 *        If CONSERVATIVE_BMI_READ is enabled, also wait for
+	 *        a BMI command credit, which indicates that the ENTIRE
+	 *        response is available in the the FIFO
+	 *
+	 *  CASE 3: length > 128
+	 *        Wait for the first 4 bytes to be in FIFO
+	 *
+	 * For most uses, a small timeout should be sufficient and we will
+	 * usually see a response quickly; but there may be some unusual
+	 * (debug) cases of BMI_EXECUTE where we want an larger timeout.
+	 * For now, we use an unbounded busy loop while waiting for
+	 * BMI_EXECUTE.
+	 *
+	 * If BMI_EXECUTE ever needs to support longer-latency execution,
+	 * especially in production, this code needs to be enhanced to sleep
+	 * and yield.  Also note that BMI_COMMUNICATION_TIMEOUT is currently
+	 * a function of Host processor speed.
+	 */
+	ret = ath10k_sdio_bmi_get_rx_lookahead(ar);
+	if (ret)
+		return ret;
+
+	/* We always read from the start of the mbox address */
+	addr = ar_sdio->mbox_info.htc_addr;
+	ret = ath10k_sdio_read_write_sync(ar, addr, ar_sdio->bmi_buf,
+					  *resp_len,
+					  HIF_RD_SYNC_BYTE_INC);
+	if (ret) {
+		ath10k_warn(ar,
+			    "unable to read the bmi data from the device: %d\n",
+			    ret);
+		return ret;
+	}
+
+	memcpy(resp, ar_sdio->bmi_buf, *resp_len);
+
+	return 0;
+}
+
+/* sdio async handling functions */
+
+static struct ath10k_sdio_bus_request
+*ath10k_sdio_alloc_busreq(struct ath10k *ar)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	struct ath10k_sdio_bus_request *bus_req;
+
+	spin_lock_bh(&ar_sdio->lock);
+
+	if (list_empty(&ar_sdio->bus_req_freeq)) {
+		bus_req = NULL;
+		goto out;
+	}
+
+	bus_req = list_first_entry(&ar_sdio->bus_req_freeq,
+				   struct ath10k_sdio_bus_request, list);
+	list_del(&bus_req->list);
+
+out:
+	spin_unlock_bh(&ar_sdio->lock);
+	return bus_req;
+}
+
+static void ath10k_sdio_free_bus_req(struct ath10k *ar,
+				     struct ath10k_sdio_bus_request *bus_req)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+
+	memset(bus_req, 0, sizeof(*bus_req));
+
+	spin_lock_bh(&ar_sdio->lock);
+	list_add_tail(&bus_req->list, &ar_sdio->bus_req_freeq);
+	spin_unlock_bh(&ar_sdio->lock);
+}
+
+static void __ath10k_sdio_write_async(struct ath10k *ar,
+				      struct ath10k_sdio_bus_request *req)
+{
+	struct ath10k_htc_ep *ep;
+	struct sk_buff *skb;
+	int ret;
+
+	skb = req->skb;
+	ret = ath10k_sdio_read_write_sync(ar, req->address,
+					  skb->data, skb->len,
+					  req->request);
+	if (req->htc_msg) {
+		ep = &ar->htc.endpoint[req->eid];
+		ath10k_htc_notify_tx_completion(ep, skb);
+	} else if (req->comp) {
+		complete(req->comp);
+	}
+
+	ath10k_sdio_free_bus_req(ar, req);
+}
+
+static void ath10k_sdio_write_async_work(struct work_struct *work)
+{
+	struct ath10k_sdio *ar_sdio = container_of(work, struct ath10k_sdio,
+						   wr_async_work);
+	struct ath10k *ar = ar_sdio->ar;
+	struct ath10k_sdio_bus_request *req, *tmp_req;
+
+	spin_lock_bh(&ar_sdio->wr_async_lock);
+
+	list_for_each_entry_safe(req, tmp_req, &ar_sdio->wr_asyncq, list) {
+		list_del(&req->list);
+		spin_unlock_bh(&ar_sdio->wr_async_lock);
+		__ath10k_sdio_write_async(ar, req);
+		spin_lock_bh(&ar_sdio->wr_async_lock);
+	}
+
+	spin_unlock_bh(&ar_sdio->wr_async_lock);
+}
+
+static int ath10k_sdio_prep_async_req(struct ath10k *ar, u32 addr,
+				      struct sk_buff *skb,
+				      struct completion *comp,
+				      bool htc_msg, enum ath10k_htc_ep_id eid)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	struct ath10k_sdio_bus_request *bus_req;
+
+	/* Allocate a bus request for the message and queue it on the
+	 * SDIO workqueue.
+	 */
+	bus_req = ath10k_sdio_alloc_busreq(ar);
+	if (!bus_req) {
+		ath10k_warn(ar,
+			    "unable to allocate bus request for async request\n");
+		return -ENOMEM;
+	}
+
+	bus_req->skb = skb;
+	bus_req->request = HIF_WRITE;
+	bus_req->eid = eid;
+	bus_req->address = addr;
+	bus_req->htc_msg = htc_msg;
+	bus_req->comp = comp;
+
+	spin_lock_bh(&ar_sdio->wr_async_lock);
+	list_add_tail(&bus_req->list, &ar_sdio->wr_asyncq);
+	spin_unlock_bh(&ar_sdio->wr_async_lock);
+
+	return 0;
+}
+
+/* IRQ handler */
+
+static void ath10k_sdio_irq_handler(struct sdio_func *func)
+{
+	struct ath10k_sdio *ar_sdio = sdio_get_drvdata(func);
+	struct ath10k *ar = ar_sdio->ar;
+	unsigned long timeout;
+	bool done = false;
+	int ret;
+
+	/* Release the host during interrupts so we can pick it back up when
+	 * we process commands.
+	 */
+	sdio_release_host(ar_sdio->func);
+
+	timeout = jiffies + ATH10K_SDIO_HIF_COMMUNICATION_TIMEOUT_HZ;
+	while (time_before(jiffies, timeout) && !done) {
+		ret = ath10k_sdio_mbox_proc_pending_irqs(ar, &done);
+		if (ret)
+			break;
+	}
+
+	sdio_claim_host(ar_sdio->func);
+
+	wake_up(&ar_sdio->irq_wq);
+
+	if (ret && ret != -ECANCELED)
+		ath10k_warn(ar, "failed to process pending SDIO interrupts: %d\n",
+			    ret);
+}
+
+/* sdio HIF functions */
+
+static int ath10k_sdio_hif_disable_intrs(struct ath10k *ar)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	struct ath10k_sdio_irq_data *irq_data = &ar_sdio->irq_data;
+	struct ath10k_sdio_irq_enable_regs *regs = irq_data->irq_en_reg;
+	int ret;
+
+	mutex_lock(&irq_data->mtx);
+
+	memset(regs, 0, sizeof(*regs));
+	ret = ath10k_sdio_read_write_sync(ar,
+					  MBOX_INT_STATUS_ENABLE_ADDRESS,
+					  &regs->int_status_en, sizeof(*regs),
+					  HIF_WR_SYNC_BYTE_INC);
+	if (ret)
+		ath10k_warn(ar, "unable to disable sdio interrupts: %d\n", ret);
+
+	mutex_unlock(&irq_data->mtx);
+
+	return ret;
+}
+
+static int ath10k_sdio_hif_power_up(struct ath10k *ar)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	struct sdio_func *func = ar_sdio->func;
+	int ret;
+
+	if (!ar_sdio->is_disabled)
+		return 0;
+
+	ath10k_dbg(ar, ATH10K_DBG_BOOT, "sdio power on\n");
+
+	sdio_claim_host(func);
+
+	ret = sdio_enable_func(func);
+	if (ret) {
+		ath10k_warn(ar, "unable to enable sdio function: %d)\n", ret);
+		sdio_release_host(func);
+		return ret;
+	}
+
+	sdio_release_host(func);
+
+	/* Wait for hardware to initialise. It should take a lot less than
+	 * 20 ms but let's be conservative here.
+	 */
+	msleep(20);
+
+	ar_sdio->is_disabled = false;
+
+	ret = ath10k_sdio_hif_disable_intrs(ar);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+static void ath10k_sdio_hif_power_down(struct ath10k *ar)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	int ret;
+
+	if (ar_sdio->is_disabled)
+		return;
+
+	ath10k_dbg(ar, ATH10K_DBG_BOOT, "sdio power off\n");
+
+	/* Disable the card */
+	sdio_claim_host(ar_sdio->func);
+	ret = sdio_disable_func(ar_sdio->func);
+	sdio_release_host(ar_sdio->func);
+
+	if (ret)
+		ath10k_warn(ar, "unable to disable sdio function: %d\n", ret);
+
+	ar_sdio->is_disabled = true;
+}
+
+static int ath10k_sdio_hif_tx_sg(struct ath10k *ar, u8 pipe_id,
+				 struct ath10k_hif_sg_item *items, int n_items)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	enum ath10k_htc_ep_id eid;
+	struct sk_buff *skb;
+	int ret, i;
+
+	eid = pipe_id_to_eid(pipe_id);
+
+	for (i = 0; i < n_items; i++) {
+		size_t padded_len;
+		u32 address;
+
+		skb = items[i].transfer_context;
+		padded_len = ath10k_sdio_calc_txrx_padded_len(ar_sdio,
+							      skb->len);
+		skb_trim(skb, padded_len);
+
+		/* Write TX data to the end of the mbox address space */
+		address = ar_sdio->mbox_addr[eid] + ar_sdio->mbox_size[eid] -
+			  skb->len;
+		ret = ath10k_sdio_prep_async_req(ar, address, skb,
+						 NULL, true, eid);
+		if (ret)
+			return ret;
+	}
+
+	queue_work(ar_sdio->workqueue, &ar_sdio->wr_async_work);
+
+	return 0;
+}
+
+static int ath10k_sdio_hif_enable_intrs(struct ath10k *ar)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	struct ath10k_sdio_irq_data *irq_data = &ar_sdio->irq_data;
+	struct ath10k_sdio_irq_enable_regs *regs = irq_data->irq_en_reg;
+	int ret;
+
+	mutex_lock(&irq_data->mtx);
+
+	/* Enable all but CPU interrupts */
+	regs->int_status_en = FIELD_PREP(MBOX_INT_STATUS_ENABLE_ERROR_MASK, 1) |
+			      FIELD_PREP(MBOX_INT_STATUS_ENABLE_CPU_MASK, 1) |
+			      FIELD_PREP(MBOX_INT_STATUS_ENABLE_COUNTER_MASK, 1);
+
+	/* NOTE: There are some cases where HIF can do detection of
+	 * pending mbox messages which is disabled now.
+	 */
+	regs->int_status_en |=
+		FIELD_PREP(MBOX_INT_STATUS_ENABLE_MBOX_DATA_MASK, 1);
+
+	/* Set up the CPU Interrupt status Register */
+	regs->cpu_int_status_en = 0;
+
+	/* Set up the Error Interrupt status Register */
+	regs->err_int_status_en =
+		FIELD_PREP(MBOX_ERROR_STATUS_ENABLE_RX_UNDERFLOW_MASK, 1) |
+		FIELD_PREP(MBOX_ERROR_STATUS_ENABLE_TX_OVERFLOW_MASK, 1);
+
+	/* Enable Counter interrupt status register to get fatal errors for
+	 * debugging.
+	 */
+	regs->cntr_int_status_en =
+		FIELD_PREP(MBOX_COUNTER_INT_STATUS_ENABLE_BIT_MASK,
+			   ATH10K_SDIO_TARGET_DEBUG_INTR_MASK);
+
+	ret = ath10k_sdio_read_write_sync(ar,
+					  MBOX_INT_STATUS_ENABLE_ADDRESS,
+					  &regs->int_status_en, sizeof(*regs),
+					  HIF_WR_SYNC_BYTE_INC);
+	if (ret)
+		ath10k_warn(ar,
+			    "failed to update mbox interrupt status register : %d\n",
+			    ret);
+
+	mutex_unlock(&irq_data->mtx);
+	return ret;
+}
+
+/* FIXME: these should be in a better place, maybe in a header file? */
+#define FIFO_TIMEOUT_AND_CHIP_CONTROL   0x00000868u
+#define FIFO_TIMEOUT_AND_CHIP_CONTROL_DISABLE_SLEEP_OFF 0xFFFEFFFF
+#define FIFO_TIMEOUT_AND_CHIP_CONTROL_DISABLE_SLEEP_ON 0x10000
+
+static int ath10k_sdio_hif_set_mbox_sleep(struct ath10k *ar, bool enable_sleep)
+{
+	u32 *val;
+	int ret;
+
+	val = kzalloc(sizeof(*val), GFP_KERNEL);
+	if (!val)
+		return -ENOMEM;
+
+	ret = ath10k_sdio_read_write_sync(ar, FIFO_TIMEOUT_AND_CHIP_CONTROL,
+					  (u8 *)val, sizeof(*val),
+					  HIF_RD_SYNC_BYTE_INC);
+	if (ret) {
+		ath10k_warn(ar, "failed to read fifo/chip control register: %d\n",
+			    ret);
+		goto out;
+	}
+
+	if (enable_sleep)
+		*val &= FIFO_TIMEOUT_AND_CHIP_CONTROL_DISABLE_SLEEP_OFF;
+	else
+		*val |= FIFO_TIMEOUT_AND_CHIP_CONTROL_DISABLE_SLEEP_ON;
+
+	ret = ath10k_sdio_read_write_sync(ar, FIFO_TIMEOUT_AND_CHIP_CONTROL,
+					  (u8 *)val, sizeof(*val),
+					  HIF_WR_SYNC_BYTE_INC);
+	if (ret)
+		goto out;
+
+	ret = 0;
+
+out:
+	kfree(val);
+
+	return ret;
+}
+
+/* set the window address register (using 4-byte register access ). */
+static int ath10k_sdio_hif_set_addrwin_reg(struct ath10k *ar, u32 reg_addr,
+					   u32 addr)
+{
+	u32 *tmp_addr;
+	int ret;
+
+	tmp_addr = kmalloc(sizeof(addr), GFP_KERNEL);
+	if (!tmp_addr)
+		return -ENOMEM;
+
+	*tmp_addr = addr;
+	ret = ath10k_sdio_read_write_sync(ar, reg_addr, (u8 *)tmp_addr,
+					  4, HIF_WR_SYNC_BYTE_INC);
+
+	if (ret) {
+		ath10k_warn(ar,
+			    "failed to write 0x%x to window reg: %d\n",
+			    reg_addr, ret);
+		goto out;
+	}
+
+	ret = 0;
+
+out:
+	kfree(tmp_addr);
+
+	return ret;
+}
+
+/* HIF diagnostics */
+
+static int ath10k_sdio_hif_diag_read(struct ath10k *ar, u32 address, void *buf,
+				     size_t buf_len)
+{
+	int ret;
+
+	/* set window register to start read cycle */
+	ret = ath10k_sdio_hif_set_addrwin_reg(ar,
+					      MBOX_WINDOW_READ_ADDR_ADDRESS,
+					      address);
+	if (ret)
+		return ret;
+
+	/* read the data */
+	ret = ath10k_sdio_read_write_sync(ar, MBOX_WINDOW_DATA_ADDRESS,
+					  (u8 *)buf, buf_len,
+					  HIF_RD_SYNC_BYTE_INC);
+	if (ret) {
+		ath10k_warn(ar, "failed to read from mbox window data addrress: %d\n",
+			    ret);
+		return ret;
+	}
+
+	return 0;
+}
+
+static int ath10k_sdio_hif_diag_read32(struct ath10k *ar, u32 address,
+				       u32 *value)
+{
+	__le32 *val;
+	int ret;
+
+	val = kzalloc(sizeof(*val), GFP_KERNEL);
+	if (!val)
+		return -ENOMEM;
+
+	ret = ath10k_sdio_hif_diag_read(ar, address, val, sizeof(*val));
+	if (ret)
+		goto out;
+
+	*value = __le32_to_cpu(*val);
+
+out:
+	kfree(val);
+
+	return ret;
+}
+
+static int ath10k_sdio_hif_diag_write_mem(struct ath10k *ar, u32 address,
+					  const void *data, int nbytes)
+{
+	int ret;
+
+	/* set write data */
+	ret = ath10k_sdio_read_write_sync(ar, MBOX_WINDOW_DATA_ADDRESS,
+					  (u8 *)data, nbytes,
+					  HIF_WR_SYNC_BYTE_INC);
+	if (ret) {
+		ath10k_warn(ar,
+			    "failed to write 0x%p to mbox window data addrress: %d\n",
+			    data, ret);
+		return ret;
+	}
+
+	/* set window register, which starts the write cycle */
+	return ath10k_sdio_hif_set_addrwin_reg(ar,
+					       MBOX_WINDOW_WRITE_ADDR_ADDRESS,
+					       address);
+}
+
+/* HIF start/stop */
+
+static int ath10k_sdio_hif_start(struct ath10k *ar)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	u32 addr, val;
+	int ret;
+
+	/* Sleep 20 ms before HIF interrupts are disabled.
+	 * This will give target plenty of time to process the BMI done
+	 * request before interrupts are disabled.
+	 */
+	msleep(20);
+	ret = ath10k_sdio_hif_disable_intrs(ar);
+	if (ret)
+		return ret;
+
+	/* eid 0 always uses the lower part of the extended mailbox address
+	 * space (ext_info[0].htc_ext_addr).
+	 */
+	ar_sdio->mbox_addr[0] = ar_sdio->mbox_info.ext_info[0].htc_ext_addr;
+	ar_sdio->mbox_size[0] = ar_sdio->mbox_info.ext_info[0].htc_ext_sz;
+
+	sdio_claim_host(ar_sdio->func);
+
+	/* Register the isr */
+	ret =  sdio_claim_irq(ar_sdio->func, ath10k_sdio_irq_handler);
+	if (ret) {
+		ath10k_warn(ar, "failed to claim sdio interrupt: %d\n", ret);
+		sdio_release_host(ar_sdio->func);
+		return ret;
+	}
+
+	sdio_release_host(ar_sdio->func);
+
+	ret = ath10k_sdio_hif_enable_intrs(ar);
+	if (ret)
+		ath10k_warn(ar, "failed to enable sdio interrupts: %d\n", ret);
+
+	addr = host_interest_item_address(HI_ITEM(hi_acs_flags));
+
+	ret = ath10k_sdio_hif_diag_read32(ar, addr, &val);
+	if (ret) {
+		ath10k_warn(ar, "unable to read hi_acs_flags address: %d\n", ret);
+		return ret;
+	}
+
+	if (val & HI_ACS_FLAGS_SDIO_SWAP_MAILBOX_FW_ACK) {
+		ath10k_dbg(ar, ATH10K_DBG_SDIO,
+			   "sdio mailbox swap service enabled\n");
+		ar_sdio->swap_mbox = true;
+	}
+
+	/* Enable sleep and then disable it again */
+	ret = ath10k_sdio_hif_set_mbox_sleep(ar, true);
+	if (ret)
+		return ret;
+
+	/* Wait for 20ms for the written value to take effect */
+	msleep(20);
+
+	ret = ath10k_sdio_hif_set_mbox_sleep(ar, false);
+	if (ret)
+		return ret;
+
+	return 0;
+}
+
+#define SDIO_IRQ_DISABLE_TIMEOUT_HZ (3 * HZ)
+
+static void ath10k_sdio_irq_disable(struct ath10k *ar)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	struct ath10k_sdio_irq_data *irq_data = &ar_sdio->irq_data;
+	struct ath10k_sdio_irq_enable_regs *regs = irq_data->irq_en_reg;
+	struct sk_buff *skb;
+	struct completion irqs_disabled_comp;
+	int ret;
+
+	skb = dev_alloc_skb(sizeof(*regs));
+	if (!skb)
+		return;
+
+	mutex_lock(&irq_data->mtx);
+
+	memset(regs, 0, sizeof(*regs)); /* disable all interrupts */
+	memcpy(skb->data, regs, sizeof(*regs));
+	skb_put(skb, sizeof(*regs));
+
+	mutex_unlock(&irq_data->mtx);
+
+	init_completion(&irqs_disabled_comp);
+	ret = ath10k_sdio_prep_async_req(ar, MBOX_INT_STATUS_ENABLE_ADDRESS,
+					 skb, &irqs_disabled_comp, false, 0);
+	if (ret)
+		goto out;
+
+	queue_work(ar_sdio->workqueue, &ar_sdio->wr_async_work);
+
+	/* Wait for the completion of the IRQ disable request.
+	 * If there is a timeout we will try to disable irq's anyway.
+	 */
+	ret = wait_for_completion_timeout(&irqs_disabled_comp,
+					  SDIO_IRQ_DISABLE_TIMEOUT_HZ);
+	if (!ret)
+		ath10k_warn(ar, "sdio irq disable request timed out\n");
+
+	sdio_claim_host(ar_sdio->func);
+
+	ret = sdio_release_irq(ar_sdio->func);
+	if (ret)
+		ath10k_warn(ar, "failed to release sdio interrupt: %d\n", ret);
+
+	sdio_release_host(ar_sdio->func);
+
+out:
+	kfree_skb(skb);
+}
+
+static void ath10k_sdio_hif_stop(struct ath10k *ar)
+{
+	struct ath10k_sdio_bus_request *req, *tmp_req;
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+
+	ath10k_sdio_irq_disable(ar);
+
+	cancel_work_sync(&ar_sdio->wr_async_work);
+
+	spin_lock_bh(&ar_sdio->wr_async_lock);
+
+	/* Free all bus requests that have not been handled */
+	list_for_each_entry_safe(req, tmp_req, &ar_sdio->wr_asyncq, list) {
+		struct ath10k_htc_ep *ep;
+
+		list_del(&req->list);
+
+		if (req->htc_msg) {
+			ep = &ar->htc.endpoint[req->eid];
+			ath10k_htc_notify_tx_completion(ep, req->skb);
+		} else if (req->skb) {
+			kfree_skb(req->skb);
+		}
+		ath10k_sdio_free_bus_req(ar, req);
+	}
+
+	spin_unlock_bh(&ar_sdio->wr_async_lock);
+}
+
+#ifdef CONFIG_PM
+
+static int ath10k_sdio_hif_suspend(struct ath10k *ar)
+{
+	return -EOPNOTSUPP;
+}
+
+static int ath10k_sdio_hif_resume(struct ath10k *ar)
+{
+	switch (ar->state) {
+	case ATH10K_STATE_OFF:
+		ath10k_dbg(ar, ATH10K_DBG_SDIO,
+			   "sdio resume configuring sdio\n");
+
+		/* need to set sdio settings after power is cut from sdio */
+		ath10k_sdio_config(ar);
+		break;
+
+	case ATH10K_STATE_ON:
+	default:
+		break;
+	}
+
+	return 0;
+}
+#endif
+
+static int ath10k_sdio_hif_map_service_to_pipe(struct ath10k *ar,
+					       u16 service_id,
+					       u8 *ul_pipe, u8 *dl_pipe)
+{
+	struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+	struct ath10k_htc *htc = &ar->htc;
+	u32 htt_addr, wmi_addr, htt_mbox_size, wmi_mbox_size;
+	enum ath10k_htc_ep_id eid;
+	bool ep_found = false;
+	int i;
+
+	/* For sdio, we are interested in the mapping between eid
+	 * and pipeid rather than service_id to pipe_id.
+	 * First we find out which eid has been allocated to the
+	 * service...
+	 */
+	for (i = 0; i < ATH10K_HTC_EP_COUNT; i++) {
+		if (htc->endpoint[i].service_id == service_id) {
+			eid = htc->endpoint[i].eid;
+			ep_found = true;
+			break;
+		}
+	}
+
+	if (!ep_found)
+		return -EINVAL;
+
+	/* Then we create the simplest mapping possible between pipeid
+	 * and eid
+	 */
+	*ul_pipe = *dl_pipe = (u8)eid;
+
+	/* Normally, HTT will use the upper part of the extended
+	 * mailbox address space (ext_info[1].htc_ext_addr) and WMI ctrl
+	 * the lower part (ext_info[0].htc_ext_addr).
+	 * If fw wants swapping of mailbox addresses, the opposite is true.
+	 */
+	if (ar_sdio->swap_mbox) {
+		htt_addr = ar_sdio->mbox_info.ext_info[0].htc_ext_addr;
+		wmi_addr = ar_sdio->mbox_info.ext_info[1].htc_ext_addr;
+		htt_mbox_size = ar_sdio->mbox_info.ext_info[0].htc_ext_sz;
+		wmi_mbox_size = ar_sdio->mbox_info.ext_info[1].htc_ext_sz;
+	} else {
+		htt_addr = ar_sdio->mbox_info.ext_info[1].htc_ext_addr;
+		wmi_addr = ar_sdio->mbox_info.ext_info[0].htc_ext_addr;
+		htt_mbox_size = ar_sdio->mbox_info.ext_info[1].htc_ext_sz;
+		wmi_mbox_size = ar_sdio->mbox_info.ext_info[0].htc_ext_sz;
+	}
+
+	switch (service_id) {
+	case ATH10K_HTC_SVC_ID_RSVD_CTRL:
+		/* HTC ctrl ep mbox address has already been setup in
+		 * ath10k_sdio_hif_start
+		 */
+		break;
+	case ATH10K_HTC_SVC_ID_WMI_CONTROL:
+		ar_sdio->mbox_addr[eid] = wmi_addr;
+		ar_sdio->mbox_size[eid] = wmi_mbox_size;
+		ath10k_dbg(ar, ATH10K_DBG_SDIO,
+			   "sdio wmi ctrl mbox_addr 0x%x mbox_size %d\n",
+			   ar_sdio->mbox_addr[eid], ar_sdio->mbox_size[eid]);
+		break;
+	case ATH10K_HTC_SVC_ID_HTT_DATA_MSG:
+		ar_sdio->mbox_addr[eid] = htt_addr;
+		ar_sdio->mbox_size[eid] = htt_mbox_size;
+		ath10k_dbg(ar, ATH10K_DBG_SDIO,
+			   "sdio htt data mbox_addr 0x%x mbox_size %d\n",
+			   ar_sdio->mbox_addr[eid], ar_sdio->mbox_size[eid]);
+		break;
+	default:
+		ath10k_warn(ar, "unsupported HTC service id: %d\n",
+			    service_id);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static void ath10k_sdio_hif_get_default_pipe(struct ath10k *ar,
+					     u8 *ul_pipe, u8 *dl_pipe)
+{
+	ath10k_dbg(ar, ATH10K_DBG_SDIO, "sdio hif get default pipe\n");
+
+	/* HTC ctrl ep (SVC id 1) always has eid (and pipe_id in our
+	 * case) == 0
+	 */
+	*ul_pipe = 0;
+	*dl_pipe = 0;
+}
+
+/* This op is currently only used by htc_wait_target if the HTC ready
+ * message times out. It is not applicable for SDIO since there is nothing
+ * we can do if the HTC ready message does not arrive in time.
+ * TODO: Make this op non mandatory by introducing a NULL check in the
+ * hif op wrapper.
+ */
+static void ath10k_sdio_hif_send_complete_check(struct ath10k *ar,
+						u8 pipe, int force)
+{
+}
+
+static const struct ath10k_hif_ops ath10k_sdio_hif_ops = {
+	.tx_sg			= ath10k_sdio_hif_tx_sg,
+	.diag_read		= ath10k_sdio_hif_diag_read,
+	.diag_write		= ath10k_sdio_hif_diag_write_mem,
+	.exchange_bmi_msg	= ath10k_sdio_bmi_exchange_msg,
+	.start			= ath10k_sdio_hif_start,
+	.stop			= ath10k_sdio_hif_stop,
+	.map_service_to_pipe	= ath10k_sdio_hif_map_service_to_pipe,
+	.get_default_pipe	= ath10k_sdio_hif_get_default_pipe,
+	.send_complete_check	= ath10k_sdio_hif_send_complete_check,
+	.power_up		= ath10k_sdio_hif_power_up,
+	.power_down		= ath10k_sdio_hif_power_down,
+#ifdef CONFIG_PM
+	.suspend		= ath10k_sdio_hif_suspend,
+	.resume			= ath10k_sdio_hif_resume,
+#endif
+};
+
+#ifdef CONFIG_PM_SLEEP
+
+/* Empty handlers so that mmc subsystem doesn't remove us entirely during
+ * suspend. We instead follow cfg80211 suspend/resume handlers.
+ */
+static int ath10k_sdio_pm_suspend(struct device *device)
+{
+	return 0;
+}
+
+static int ath10k_sdio_pm_resume(struct device *device)
+{
+	return 0;
+}
+
+static SIMPLE_DEV_PM_OPS(ath10k_sdio_pm_ops, ath10k_sdio_pm_suspend,
+			 ath10k_sdio_pm_resume);
+
+#define ATH10K_SDIO_PM_OPS (&ath10k_sdio_pm_ops)
+
+#else
+
+#define ATH10K_SDIO_PM_OPS NULL
+
+#endif /* CONFIG_PM_SLEEP */
+
+static int ath10k_sdio_probe(struct sdio_func *func,
+			     const struct sdio_device_id *id)
+{
+	struct ath10k_sdio *ar_sdio;
+	struct ath10k *ar;
+	enum ath10k_hw_rev hw_rev;
+	u32 chip_id, dev_id_base;
+	int ret, i;
+
+	/* Assumption: All SDIO based chipsets (so far) are QCA6174 based.
+	 * If there will be newer chipsets that does not use the hw reg
+	 * setup as defined in qca6174_regs and qca6174_values, this
+	 * assumption is no longer valid and hw_rev must be setup differently
+	 * depending on chipset.
+	 */
+	hw_rev = ATH10K_HW_QCA6174;
+
+	ar = ath10k_core_create(sizeof(*ar_sdio), &func->dev, ATH10K_BUS_SDIO,
+				hw_rev, &ath10k_sdio_hif_ops);
+	if (!ar) {
+		dev_err(&func->dev, "failed to allocate core\n");
+		return -ENOMEM;
+	}
+
+	ath10k_dbg(ar, ATH10K_DBG_BOOT,
+		   "sdio new func %d vendor 0x%x device 0x%x block 0x%x/0x%x\n",
+		   func->num, func->vendor, func->device,
+		   func->max_blksize, func->cur_blksize);
+
+	ar_sdio = ath10k_sdio_priv(ar);
+
+	ar_sdio->irq_data.irq_proc_reg =
+		kzalloc(sizeof(struct ath10k_sdio_irq_proc_regs),
+			GFP_KERNEL);
+	if (!ar_sdio->irq_data.irq_proc_reg) {
+		ret = -ENOMEM;
+		goto err_core_destroy;
+	}
+
+	ar_sdio->irq_data.irq_en_reg =
+		kzalloc(sizeof(struct ath10k_sdio_irq_enable_regs),
+			GFP_KERNEL);
+	if (!ar_sdio->irq_data.irq_en_reg) {
+		ret = -ENOMEM;
+		goto err_free_proc_reg;
+	}
+
+	ar_sdio->bmi_buf = kzalloc(BMI_MAX_CMDBUF_SIZE, GFP_KERNEL);
+	if (!ar_sdio->bmi_buf) {
+		ret = -ENOMEM;
+		goto err_free_en_reg;
+	}
+
+	ar_sdio->func = func;
+	sdio_set_drvdata(func, ar_sdio);
+
+	ar_sdio->is_disabled = true;
+	ar_sdio->ar = ar;
+
+	spin_lock_init(&ar_sdio->lock);
+	spin_lock_init(&ar_sdio->wr_async_lock);
+	mutex_init(&ar_sdio->irq_data.mtx);
+
+	INIT_LIST_HEAD(&ar_sdio->bus_req_freeq);
+	INIT_LIST_HEAD(&ar_sdio->wr_asyncq);
+
+	INIT_WORK(&ar_sdio->wr_async_work, ath10k_sdio_write_async_work);
+	ar_sdio->workqueue = create_singlethread_workqueue("ath10k_sdio_wq");
+	if (!ar_sdio->workqueue) {
+		ret = -ENOMEM;
+		goto err_free_bmi_buf;
+	}
+
+	init_waitqueue_head(&ar_sdio->irq_wq);
+
+	for (i = 0; i < ATH10K_SDIO_BUS_REQUEST_MAX_NUM; i++)
+		ath10k_sdio_free_bus_req(ar, &ar_sdio->bus_req[i]);
+
+	dev_id_base = FIELD_GET(QCA_MANUFACTURER_ID_BASE, id->device);
+	switch (dev_id_base) {
+	case QCA_MANUFACTURER_ID_AR6005_BASE:
+	case QCA_MANUFACTURER_ID_QCA9377_BASE:
+		ar->dev_id = QCA9377_1_0_DEVICE_ID;
+		break;
+	default:
+		ret = -ENODEV;
+		ath10k_err(ar, "unsupported device id %u (0x%x)\n",
+			   dev_id_base, id->device);
+		goto err_free_bmi_buf;
+	}
+
+	ar->id.vendor = id->vendor;
+	ar->id.device = id->device;
+
+	ath10k_sdio_set_mbox_info(ar);
+
+	ret = ath10k_sdio_config(ar);
+	if (ret) {
+		ath10k_err(ar, "failed to config sdio: %d\n", ret);
+		goto err_free_wq;
+	}
+
+	/* TODO: don't know yet how to get chip_id with SDIO */
+	chip_id = 0;
+	ret = ath10k_core_register(ar, chip_id);
+	if (ret) {
+		ath10k_err(ar, "failed to register driver core: %d\n", ret);
+		goto err_free_wq;
+	}
+
+	/* TODO: remove this once SDIO support is fully implemented */
+	ath10k_warn(ar, "WARNING: ath10k SDIO support is incomplete, don't expect anything to work!\n");
+
+	return 0;
+
+err_free_wq:
+	destroy_workqueue(ar_sdio->workqueue);
+err_free_bmi_buf:
+	kfree(ar_sdio->bmi_buf);
+err_free_en_reg:
+	kfree(ar_sdio->irq_data.irq_en_reg);
+err_free_proc_reg:
+	kfree(ar_sdio->irq_data.irq_proc_reg);
+err_core_destroy:
+	ath10k_core_destroy(ar);
+
+	return ret;
+}
+
+static void ath10k_sdio_remove(struct sdio_func *func)
+{
+	struct ath10k_sdio *ar_sdio = sdio_get_drvdata(func);
+	struct ath10k *ar = ar_sdio->ar;
+
+	ath10k_dbg(ar, ATH10K_DBG_BOOT,
+		   "sdio removed func %d vendor 0x%x device 0x%x\n",
+		   func->num, func->vendor, func->device);
+
+	(void)ath10k_sdio_hif_disable_intrs(ar);
+	cancel_work_sync(&ar_sdio->wr_async_work);
+	ath10k_core_unregister(ar);
+	ath10k_core_destroy(ar);
+}
+
+static const struct sdio_device_id ath10k_sdio_devices[] = {
+	{SDIO_DEVICE(QCA_MANUFACTURER_CODE,
+		     (QCA_SDIO_ID_AR6005_BASE | 0xA))},
+	{SDIO_DEVICE(QCA_MANUFACTURER_CODE,
+		     (QCA_SDIO_ID_QCA9377_BASE | 0x1))},
+	{},
+};
+
+MODULE_DEVICE_TABLE(sdio, ath10k_sdio_devices);
+
+static struct sdio_driver ath10k_sdio_driver = {
+	.name = "ath10k_sdio",
+	.id_table = ath10k_sdio_devices,
+	.probe = ath10k_sdio_probe,
+	.remove = ath10k_sdio_remove,
+	.drv.pm = ATH10K_SDIO_PM_OPS,
+};
+
+static int __init ath10k_sdio_init(void)
+{
+	int ret;
+
+	ret = sdio_register_driver(&ath10k_sdio_driver);
+	if (ret)
+		pr_err("sdio driver registration failed: %d\n", ret);
+
+	return ret;
+}
+
+static void __exit ath10k_sdio_exit(void)
+{
+	sdio_unregister_driver(&ath10k_sdio_driver);
+}
+
+module_init(ath10k_sdio_init);
+module_exit(ath10k_sdio_exit);
+
+MODULE_AUTHOR("Qualcomm Atheros");
+MODULE_DESCRIPTION("Driver support for Qualcomm Atheros 802.11ac WLAN SDIO devices");
+MODULE_LICENSE("Dual BSD/GPL");
diff --git a/drivers/net/wireless/ath/ath10k/sdio.h b/drivers/net/wireless/ath/ath10k/sdio.h
new file mode 100644
index 000000000000..bc8d6e5342ea
--- /dev/null
+++ b/drivers/net/wireless/ath/ath10k/sdio.h
@@ -0,0 +1,263 @@ 
+/*
+ * Copyright (c) 2004-2011 Atheros Communications Inc.
+ * Copyright (c) 2011-2012 Qualcomm Atheros, Inc.
+ * Copyright (c) 2016-2017 Erik Stromdahl <erik.stromdahl@gmail.com>
+ *
+ * Permission to use, copy, modify, and/or distribute this software for any
+ * purpose with or without fee is hereby granted, provided that the above
+ * copyright notice and this permission notice appear in all copies.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+ * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+ * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+ * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#ifndef _SDIO_H_
+#define _SDIO_H_
+
+#define ATH10K_HIF_MBOX_BLOCK_SIZE              256
+
+#define QCA_MANUFACTURER_ID_BASE                GENMASK(11, 8)
+#define QCA_MANUFACTURER_ID_AR6005_BASE         0x5
+#define QCA_MANUFACTURER_ID_QCA9377_BASE        0x7
+#define QCA_SDIO_ID_AR6005_BASE                 0x500
+#define QCA_SDIO_ID_QCA9377_BASE                0x700
+#define QCA_MANUFACTURER_ID_REV_MASK            0x00FF
+#define QCA_MANUFACTURER_CODE                   0x271 /* Qualcomm/Atheros */
+
+#define ATH10K_SDIO_MAX_BUFFER_SIZE             4096 /*Unsure of this constant*/
+
+/* Mailbox address in SDIO address space */
+#define ATH10K_HIF_MBOX_BASE_ADDR               0x1000
+#define ATH10K_HIF_MBOX_WIDTH                   0x800
+
+#define ATH10K_HIF_MBOX_TOT_WIDTH \
+	(ATH10K_HIF_MBOX_NUM_MAX * ATH10K_HIF_MBOX_WIDTH)
+
+#define ATH10K_HIF_MBOX0_EXT_BASE_ADDR          0x5000
+#define ATH10K_HIF_MBOX0_EXT_WIDTH              (36 * 1024)
+#define ATH10K_HIF_MBOX0_EXT_WIDTH_ROME_2_0     (56 * 1024)
+#define ATH10K_HIF_MBOX1_EXT_WIDTH              (36 * 1024)
+#define ATH10K_HIF_MBOX_DUMMY_SPACE_SIZE        (2 * 1024)
+
+#define ATH10K_HTC_MBOX_MAX_PAYLOAD_LENGTH \
+	(ATH10K_SDIO_MAX_BUFFER_SIZE - sizeof(struct ath10k_htc_hdr))
+
+#define ATH10K_HIF_MBOX_NUM_MAX                 4
+#define ATH10K_SDIO_BUS_REQUEST_MAX_NUM         64
+
+#define ATH10K_SDIO_HIF_COMMUNICATION_TIMEOUT_HZ (100 * HZ)
+
+/* HTC runs over mailbox 0 */
+#define ATH10K_HTC_MAILBOX                      0
+#define ATH10K_HTC_MAILBOX_MASK                 BIT(ATH10K_HTC_MAILBOX)
+
+/* GMBOX addresses */
+#define ATH10K_HIF_GMBOX_BASE_ADDR              0x7000
+#define ATH10K_HIF_GMBOX_WIDTH                  0x4000
+
+/* Modified versions of the sdio.h macros.
+ * The macros in sdio.h can't be used easily with the FIELD_{PREP|GET}
+ * macros in bitfield.h, so we define our own macros here.
+ */
+#define ATH10K_SDIO_DRIVE_DTSX_MASK \
+	(SDIO_DRIVE_DTSx_MASK << SDIO_DRIVE_DTSx_SHIFT)
+
+#define ATH10K_SDIO_DRIVE_DTSX_TYPE_B           0
+#define ATH10K_SDIO_DRIVE_DTSX_TYPE_A           1
+#define ATH10K_SDIO_DRIVE_DTSX_TYPE_C           2
+#define ATH10K_SDIO_DRIVE_DTSX_TYPE_D           3
+
+/* SDIO CCCR register definitions */
+#define CCCR_SDIO_IRQ_MODE_REG                  0xF0
+#define CCCR_SDIO_IRQ_MODE_REG_SDIO3            0x16
+
+#define CCCR_SDIO_DRIVER_STRENGTH_ENABLE_ADDR   0xF2
+
+#define CCCR_SDIO_DRIVER_STRENGTH_ENABLE_A      0x02
+#define CCCR_SDIO_DRIVER_STRENGTH_ENABLE_C      0x04
+#define CCCR_SDIO_DRIVER_STRENGTH_ENABLE_D      0x08
+
+#define CCCR_SDIO_ASYNC_INT_DELAY_ADDRESS       0xF0
+#define CCCR_SDIO_ASYNC_INT_DELAY_MASK          0xC0
+
+/* mode to enable special 4-bit interrupt assertion without clock */
+#define SDIO_IRQ_MODE_ASYNC_4BIT_IRQ            BIT(0)
+#define SDIO_IRQ_MODE_ASYNC_4BIT_IRQ_SDIO3      BIT(1)
+
+#define ATH10K_SDIO_TARGET_DEBUG_INTR_MASK      0x01
+
+/* The theoretical maximum number of RX messages that can be fetched
+ * from the mbox interrupt handler in one loop is derived in the following
+ * way:
+ *
+ * Let's assume that each packet in a bundle of the maximum bundle size
+ * (HTC_HOST_MAX_MSG_PER_BUNDLE) has the HTC header bundle count set
+ * to the maximum value (HTC_HOST_MAX_MSG_PER_BUNDLE).
+ *
+ * in this case the driver must allocate
+ * (HTC_HOST_MAX_MSG_PER_BUNDLE * HTC_HOST_MAX_MSG_PER_BUNDLE) skb's.
+ */
+#define ATH10K_SDIO_MAX_RX_MSGS \
+	(HTC_HOST_MAX_MSG_PER_BUNDLE * HTC_HOST_MAX_MSG_PER_BUNDLE)
+
+struct ath10k_sdio_bus_request {
+	struct list_head list;
+
+	/* sdio address */
+	u32 address;
+	u32 request;
+	struct sk_buff *skb;
+	enum ath10k_htc_ep_id eid;
+	int status;
+	/* Specifies if the current request is an HTC message.
+	 * If not, the eid is not applicable an the TX completion handler
+	 * associated with the endpoint will not be invoked.
+	 */
+	bool htc_msg;
+	/* Completion that (if set) will be invoked for non HTC requests
+	 * (htc_msg == false) when the request has been processed.
+	 */
+	struct completion *comp;
+};
+
+struct ath10k_sdio_rx_data {
+	struct sk_buff *skb;
+	size_t alloc_len;
+	size_t act_len;
+	enum ath10k_htc_ep_id eid;
+	bool part_of_bundle;
+	bool last_in_bundle;
+	bool trailer_only;
+	int status;
+};
+
+/* direction of transfer (read/write) */
+#define HIF_READ                    0x00000001
+#define HIF_WRITE                   0x00000002
+
+/*     dmode - An interface may support different kinds of commands based on
+ *             the tradeoff between the amount of data it can carry and the
+ *             setup time. Byte and Block modes are supported (HIF_BYTE_BASIS/
+ *             HIF_BLOCK_BASIS). In case of latter, the data is rounded off
+ *             to the nearest block size by padding. The size of the block is
+ *             configurable at compile time using the HIF_BLOCK_SIZE and is
+ *             negotiated with the target during initialization after the
+ *             ATH10K interrupts are enabled.
+ */
+#define HIF_BYTE_BASIS              0x00000040
+#define HIF_BLOCK_BASIS             0x00000080
+
+/*     amode - This indicates if the address has to be incremented on ATH10K
+ *             after every read/write operation (HIF?FIXED_ADDRESS/
+ *             HIF_INCREMENTAL_ADDRESS).
+ */
+#define HIF_FIXED_ADDRESS           0x00000100
+#define HIF_INCREMENTAL_ADDRESS     0x00000200
+
+#define HIF_WR_SYNC_BYTE_FIX					\
+	(HIF_WRITE | HIF_BYTE_BASIS | HIF_FIXED_ADDRESS)
+
+#define HIF_WR_SYNC_BYTE_INC					\
+	(HIF_WRITE | HIF_BYTE_BASIS | HIF_INCREMENTAL_ADDRESS)
+
+#define HIF_RD_SYNC_BYTE_INC						\
+	(HIF_READ | HIF_BYTE_BASIS | HIF_INCREMENTAL_ADDRESS)
+
+#define HIF_RD_SYNC_BYTE_FIX						\
+	(HIF_READ | HIF_BYTE_BASIS | HIF_FIXED_ADDRESS)
+
+#define HIF_RD_SYNC_BLOCK_FIX						\
+	(HIF_READ | HIF_BLOCK_BASIS | HIF_FIXED_ADDRESS)
+
+struct ath10k_sdio_irq_proc_regs {
+	u8 host_int_status;
+	u8 cpu_int_status;
+	u8 error_int_status;
+	u8 counter_int_status;
+	u8 mbox_frame;
+	u8 rx_lookahead_valid;
+	u8 host_int_status2;
+	u8 gmbox_rx_avail;
+	__le32 rx_lookahead[2];
+	__le32 rx_gmbox_lookahead_alias[2];
+};
+
+struct ath10k_sdio_irq_enable_regs {
+	u8 int_status_en;
+	u8 cpu_int_status_en;
+	u8 err_int_status_en;
+	u8 cntr_int_status_en;
+};
+
+struct ath10k_sdio_irq_data {
+	/* protects irq_proc_reg and irq_en_reg below.
+	 * We use a mutex here and not a spinlock since we will have the
+	 * mutex locked while calling the sdio_memcpy_ functions.
+	 * These function require non atomic context, and hence, spinlocks
+	 * can be held while calling these functions.
+	 */
+	struct mutex mtx;
+	struct ath10k_sdio_irq_proc_regs *irq_proc_reg;
+	struct ath10k_sdio_irq_enable_regs *irq_en_reg;
+};
+
+struct ath10k_mbox_ext_info {
+	u32 htc_ext_addr;
+	u32 htc_ext_sz;
+};
+
+struct ath10k_mbox_info {
+	u32 htc_addr;
+	struct ath10k_mbox_ext_info ext_info[2];
+	u32 block_size;
+	u32 block_mask;
+	u32 gmbox_addr;
+	u32 gmbox_sz;
+};
+
+struct ath10k_sdio {
+	struct sdio_func *func;
+
+	struct ath10k_mbox_info mbox_info;
+	bool swap_mbox;
+	u32 mbox_addr[ATH10K_HTC_EP_COUNT];
+	u32 mbox_size[ATH10K_HTC_EP_COUNT];
+
+	/* available bus requests */
+	struct ath10k_sdio_bus_request bus_req[ATH10K_SDIO_BUS_REQUEST_MAX_NUM];
+	/* free list of bus requests */
+	struct list_head bus_req_freeq;
+	/* protects access to bus_req_freeq */
+	spinlock_t lock;
+
+	struct ath10k_sdio_rx_data rx_pkts[ATH10K_SDIO_MAX_RX_MSGS];
+	size_t n_rx_pkts;
+
+	struct ath10k *ar;
+	struct ath10k_sdio_irq_data irq_data;
+
+	/* temporary buffer for BMI requests */
+	u8 *bmi_buf;
+
+	wait_queue_head_t irq_wq;
+
+	bool is_disabled;
+
+	struct workqueue_struct *workqueue;
+	struct work_struct wr_async_work;
+	struct list_head wr_asyncq;
+	/* protects access to wr_asyncq */
+	spinlock_t wr_async_lock;
+};
+
+static inline struct ath10k_sdio *ath10k_sdio_priv(struct ath10k *ar)
+{
+	return (struct ath10k_sdio *)ar->drv_priv;
+}
+
+#endif