ath10k: add initial SDIO support
authorErik Stromdahl <erik.stromdahl@gmail.com>
Wed, 26 Apr 2017 09:18:00 +0000 (12:18 +0300)
committerKalle Valo <kvalo@qca.qualcomm.com>
Thu, 4 May 2017 12:55:55 +0000 (15:55 +0300)
Chipsets like QCA6584 have support for SDIO so add initial SDIO bus support to
ath10k. With this patch we have the low level HTC protocol working and it's
possible to boot the firmware, but it's still not possible to connect or
anything like. More changes are needed for full functionality. For that reason
we print during initialisation:

WARNING: ath10k SDIO support is incomplete, don't expect anything to work!

Signed-off-by: Erik Stromdahl <erik.stromdahl@gmail.com>
[kvalo@qca.qualcomm.com: refactoring, cleanup, commit log]
Signed-off-by: Kalle Valo <kvalo@qca.qualcomm.com>
drivers/net/wireless/ath/ath10k/Kconfig
drivers/net/wireless/ath/ath10k/Makefile
drivers/net/wireless/ath/ath10k/sdio.c [new file with mode: 0644]
drivers/net/wireless/ath/ath10k/sdio.h [new file with mode: 0644]

index b4241cf..412eb13 100644 (file)
@@ -22,6 +22,13 @@ 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 experimental support for SDIO/MMC bus. Currently
+         work in progress and will not fully work.
+
 config ATH10K_DEBUG
        bool "Atheros ath10k debugging"
        depends on ATH10K
index 930fadd..b0b19a7 100644 (file)
@@ -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 (file)
index 0000000..9e78fba
--- /dev/null
@@ -0,0 +1,2113 @@
+/*
+ * Copyright (c) 2004-2011 Atheros Communications Inc.
+ * Copyright (c) 2011-2012,2017 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_write32(struct ath10k *ar, u32 addr, u32 val)
+{
+       struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+       struct sdio_func *func = ar_sdio->func;
+       int ret;
+
+       sdio_claim_host(func);
+
+       sdio_writel(func, val, addr, &ret);
+       if (ret) {
+               ath10k_warn(ar, "failed to write 0x%x to address 0x%x: %d\n",
+                           val, addr, ret);
+               goto out;
+       }
+
+       ath10k_dbg(ar, ATH10K_DBG_SDIO, "sdio write32 addr 0x%x val 0x%x\n",
+                  addr, val);
+
+out:
+       sdio_release_host(func);
+
+       return ret;
+}
+
+static int ath10k_sdio_writesb32(struct ath10k *ar, u32 addr, u32 val)
+{
+       struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+       struct sdio_func *func = ar_sdio->func;
+       __le32 *buf;
+       int ret;
+
+       buf = kzalloc(sizeof(*buf), GFP_KERNEL);
+       if (!buf)
+               return -ENOMEM;
+
+       *buf = cpu_to_le32(val);
+
+       sdio_claim_host(func);
+
+       ret = sdio_writesb(func, addr, buf, sizeof(*buf));
+       if (ret) {
+               ath10k_warn(ar, "failed to write value 0x%x to fixed sb address 0x%x: %d\n",
+                           val, addr, ret);
+               goto out;
+       }
+
+       ath10k_dbg(ar, ATH10K_DBG_SDIO, "sdio writesb32 addr 0x%x val 0x%x\n",
+                  addr, val);
+
+out:
+       sdio_release_host(func);
+
+       kfree(buf);
+
+       return ret;
+}
+
+static int ath10k_sdio_read32(struct ath10k *ar, u32 addr, u32 *val)
+{
+       struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+       struct sdio_func *func = ar_sdio->func;
+       int ret;
+
+       sdio_claim_host(func);
+       *val = sdio_readl(func, addr, &ret);
+       if (ret) {
+               ath10k_warn(ar, "failed to read from address 0x%x: %d\n",
+                           addr, ret);
+               goto out;
+       }
+
+       ath10k_dbg(ar, ATH10K_DBG_SDIO, "sdio read32 addr 0x%x val 0x%x\n",
+                  addr, *val);
+
+out:
+       sdio_release_host(func);
+
+       return ret;
+}
+
+static int ath10k_sdio_read(struct ath10k *ar, u32 addr, void *buf, size_t len)
+{
+       struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+       struct sdio_func *func = ar_sdio->func;
+       int ret;
+
+       sdio_claim_host(func);
+
+       ret = sdio_memcpy_fromio(func, buf, addr, len);
+       if (ret) {
+               ath10k_warn(ar, "failed to read from address 0x%x: %d\n",
+                           addr, ret);
+               goto out;
+       }
+
+       ath10k_dbg(ar, ATH10K_DBG_SDIO, "sdio read addr 0x%x buf 0x%p len %zu\n",
+                  addr, buf, len);
+       ath10k_dbg_dump(ar, ATH10K_DBG_SDIO_DUMP, NULL, "sdio read ", buf, len);
+
+out:
+       sdio_release_host(func);
+
+       return ret;
+}
+
+static int ath10k_sdio_write(struct ath10k *ar, u32 addr, const void *buf, size_t len)
+{
+       struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+       struct sdio_func *func = ar_sdio->func;
+       int ret;
+
+       sdio_claim_host(func);
+
+       /* For some reason toio() doesn't have const for the buffer, need
+        * an ugly hack to workaround that.
+        */
+       ret = sdio_memcpy_toio(func, addr, (void *)buf, len);
+       if (ret) {
+               ath10k_warn(ar, "failed to write to address 0x%x: %d\n",
+                           addr, ret);
+               goto out;
+       }
+
+       ath10k_dbg(ar, ATH10K_DBG_SDIO, "sdio write addr 0x%x buf 0x%p len %zu\n",
+                  addr, buf, len);
+       ath10k_dbg_dump(ar, ATH10K_DBG_SDIO_DUMP, NULL, "sdio write ", buf, len);
+
+out:
+       sdio_release_host(func);
+
+       return ret;
+}
+
+static int ath10k_sdio_readsb(struct ath10k *ar, u32 addr, void *buf, size_t len)
+{
+       struct ath10k_sdio *ar_sdio = ath10k_sdio_priv(ar);
+       struct sdio_func *func = ar_sdio->func;
+       int ret;
+
+       sdio_claim_host(func);
+
+       len = round_down(len, ar_sdio->mbox_info.block_size);
+
+       ret = sdio_readsb(func, buf, addr, len);
+       if (ret) {
+               ath10k_warn(ar, "failed to read from fixed (sb) address 0x%x: %d\n",
+                           addr, ret);
+               goto out;
+       }
+
+       ath10k_dbg(ar, ATH10K_DBG_SDIO, "sdio readsb addr 0x%x buf 0x%p len %zu\n",
+                  addr, buf, len);
+       ath10k_dbg_dump(ar, ATH10K_DBG_SDIO_DUMP, NULL, "sdio readsb ", buf, len);
+
+out:
+       sdio_release_host(func);
+
+       return ret;
+}
+
+/* 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_readsb(ar, ar_sdio->mbox_info.htc_addr,
+                                skb->data, pkt->alloc_len);
+       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 val;
+       int ret;
+
+       /* 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_read32(ar, MBOX_COUNT_DEC_ADDRESS, &val);
+       if (ret)
+               ath10k_warn(ar, "failed to clear debug interrupt: %d\n", ret);
+
+       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;
+       int ret;
+
+       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);
+               return -EIO;
+       }
+
+       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 */
+       ret = ath10k_sdio_writesb32(ar, MBOX_ERROR_INT_STATUS_ADDRESS,
+                                   error_int_status);
+       if (ret) {
+               ath10k_warn(ar, "unable to write to error int status address: %d\n",
+                           ret);
+               return ret;
+       }
+
+       return 0;
+}
+
+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;
+       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");
+               ret = -EIO;
+               goto out;
+       }
+
+       /* 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.
+        *
+        * Set W1C value to clear the interrupt, this hits the register first.
+        */
+       ret = ath10k_sdio_writesb32(ar, MBOX_CPU_INT_STATUS_ADDRESS,
+                                   cpu_int_status);
+       if (ret) {
+               ath10k_warn(ar, "unable to write to cpu interrupt status address: %d\n",
+                           ret);
+               goto out;
+       }
+
+out:
+       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(ar, MBOX_HOST_INT_STATUS_ADDRESS,
+                              irq_proc_reg, sizeof(*irq_proc_reg));
+       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;
+               ret = 0;
+               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)
+{
+       u32 addr, cmd_credits;
+       unsigned long timeout;
+       int ret;
+
+       /* 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;
+       cmd_credits = 0;
+
+       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_read32(ar, addr, &cmd_credits);
+               if (ret) {
+                       ath10k_warn(ar,
+                                   "unable to decrement the command credit count register: %d\n",
+                                   ret);
+                       return ret;
+               }
+
+               /* 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");
+               return -ETIMEDOUT;
+       }
+
+       return 0;
+}
+
+static int ath10k_sdio_bmi_get_rx_lookahead(struct ath10k *ar)
+{
+       unsigned long timeout;
+       u32 rx_word;
+       int ret;
+
+       timeout = jiffies + BMI_COMMUNICATION_TIMEOUT_HZ;
+       rx_word = 0;
+
+       while ((time_before(jiffies, timeout)) && !rx_word) {
+               ret = ath10k_sdio_read32(ar,
+                                        MBOX_HOST_INT_STATUS_ADDRESS,
+                                        &rx_word);
+               if (ret) {
+                       ath10k_warn(ar, "unable to read RX_LOOKAHEAD_VALID: %d\n", ret);
+                       return ret;
+               }
+
+                /* all we really want is one bit */
+               rx_word &= 1;
+       }
+
+       if (!rx_word) {
+               ath10k_warn(ar, "bmi_recv_buf FIFO empty\n");
+               return -EINVAL;
+       }
+
+       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_write(ar, addr, ar_sdio->bmi_buf, req_len);
+               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(ar, addr, ar_sdio->bmi_buf, *resp_len);
+       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_write(ar, req->address, skb->data, skb->len);
+       if (ret)
+               ath10k_warn(ar, "failed to write skb to 0x%x asynchronously: %d",
+                           req->address, ret);
+
+       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->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_write(ar, MBOX_INT_STATUS_ENABLE_ADDRESS,
+                               &regs->int_status_en, sizeof(*regs));
+       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_write(ar, MBOX_INT_STATUS_ENABLE_ADDRESS,
+                               &regs->int_status_en, sizeof(*regs));
+       if (ret)
+               ath10k_warn(ar,
+                           "failed to update mbox interrupt status register : %d\n",
+                           ret);
+
+       mutex_unlock(&irq_data->mtx);
+       return ret;
+}
+
+static int ath10k_sdio_hif_set_mbox_sleep(struct ath10k *ar, bool enable_sleep)
+{
+       u32 val;
+       int ret;
+
+       ret = ath10k_sdio_read32(ar, ATH10K_FIFO_TIMEOUT_AND_CHIP_CONTROL, &val);
+       if (ret) {
+               ath10k_warn(ar, "failed to read fifo/chip control register: %d\n",
+                           ret);
+               return ret;
+       }
+
+       if (enable_sleep)
+               val &= ATH10K_FIFO_TIMEOUT_AND_CHIP_CONTROL_DISABLE_SLEEP_OFF;
+       else
+               val |= ATH10K_FIFO_TIMEOUT_AND_CHIP_CONTROL_DISABLE_SLEEP_ON;
+
+       ret = ath10k_sdio_write32(ar, ATH10K_FIFO_TIMEOUT_AND_CHIP_CONTROL, val);
+       if (ret) {
+               ath10k_warn(ar, "failed to write to FIFO_TIMEOUT_AND_CHIP_CONTROL: %d",
+                           ret);
+               return ret;
+       }
+
+       return 0;
+}
+
+/* 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_write32(ar, MBOX_WINDOW_READ_ADDR_ADDRESS, address);
+       if (ret) {
+               ath10k_warn(ar, "failed to set mbox window read address: %d", ret);
+               return ret;
+       }
+
+       /* read the data */
+       ret = ath10k_sdio_read(ar, MBOX_WINDOW_DATA_ADDRESS, buf, buf_len);
+       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_write(ar, MBOX_WINDOW_DATA_ADDRESS, data, nbytes);
+       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 */
+       ret = ath10k_sdio_write32(ar, MBOX_WINDOW_WRITE_ADDR_ADDRESS, address);
+       if (ret) {
+               ath10k_warn(ar, "failed to set mbox window write address: %d", ret);
+               return ret;
+       }
+
+       return 0;
+}
+
+/* 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 (file)
index 0000000..3f61c67
--- /dev/null
@@ -0,0 +1,229 @@
+/*
+ * 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)
+
+#define ATH10K_FIFO_TIMEOUT_AND_CHIP_CONTROL   0x00000868u
+#define ATH10K_FIFO_TIMEOUT_AND_CHIP_CONTROL_DISABLE_SLEEP_OFF 0xFFFEFFFF
+#define ATH10K_FIFO_TIMEOUT_AND_CHIP_CONTROL_DISABLE_SLEEP_ON 0x10000
+
+struct ath10k_sdio_bus_request {
+       struct list_head list;
+
+       /* sdio address */
+       u32 address;
+
+       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;
+};
+
+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