OSDN Git Service

net: wwan: Add Qualcomm BAM-DMUX WWAN network driver
authorStephan Gerhold <stephan@gerhold.net>
Sat, 27 Nov 2021 17:31:08 +0000 (18:31 +0100)
committerDavid S. Miller <davem@davemloft.net>
Mon, 29 Nov 2021 12:27:34 +0000 (12:27 +0000)
The BAM Data Multiplexer provides access to the network data channels of
modems integrated into many older Qualcomm SoCs, e.g. Qualcomm MSM8916 or
MSM8974. It is built using a simple protocol layer on top of a DMA engine
(Qualcomm BAM) and bidirectional interrupts to coordinate power control.

The modem announces a fixed set of channels by sending an OPEN command.
The driver exports each channel as separate network interface so that
a connection can be established via QMI from userspace. The network
interface can work either in Ethernet or Raw-IP mode (configurable via
QMI). However, Ethernet mode seems to be broken with most firmwares
(network packets are actually received as Raw-IP), therefore the driver
only supports Raw-IP mode.

Note that the control channel (QMI/AT) is entirely separate from
BAM-DMUX and is already supported by the RPMSG_WWAN_CTRL driver.

The driver uses runtime PM to coordinate power control with the modem.
TX/RX buffers are put in a kind of "ring queue" and submitted via
the bam_dma driver of the DMAEngine subsystem.

The basic architecture looks roughly like this:

                   +------------+                +-------+
         [IPv4/6]  |  BAM-DMUX  |                |       |
         [Data...] |            |                |       |
        ---------->|wwan0       | [DMUX chan: x] |       |
         [IPv4/6]  | (chan: 0)  | [IPv4/6]       |       |
         [Data...] |            | [Data...]      |       |
        ---------->|wwan1       |--------------->| Modem |
                   | (chan: 1)  |      BAM       |       |
         [IPv4/6]  | ...        |  (DMA Engine)  |       |
         [Data...] |            |                |       |
        ---------->|wwan7       |                |       |
                   | (chan: 7)  |                |       |
                   +------------+                +-------+

Note that some newer firmware versions support QMAP ("rmnet" driver)
as additional multiplexing layer on top of BAM-DMUX, but this is not
currently supported by this driver.

Signed-off-by: Stephan Gerhold <stephan@gerhold.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
MAINTAINERS
drivers/net/wwan/Kconfig
drivers/net/wwan/Makefile
drivers/net/wwan/qcom_bam_dmux.c [new file with mode: 0644]

index 360e9aa..167d989 100644 (file)
@@ -15727,6 +15727,14 @@ W:     https://wireless.wiki.kernel.org/en/users/Drivers/ath9k
 F:     Documentation/devicetree/bindings/net/wireless/qca,ath9k.yaml
 F:     drivers/net/wireless/ath/ath9k/
 
+QUALCOMM BAM-DMUX WWAN NETWORK DRIVER
+M:     Stephan Gerhold <stephan@gerhold.net>
+L:     netdev@vger.kernel.org
+L:     linux-arm-msm@vger.kernel.org
+S:     Maintained
+F:     Documentation/devicetree/bindings/net/qcom,bam-dmux.yaml
+F:     drivers/net/wwan/qcom_bam_dmux.c
+
 QUALCOMM CAMERA SUBSYSTEM DRIVER
 M:     Robert Foss <robert.foss@linaro.org>
 M:     Todor Tomov <todor.too@gmail.com>
index 17543be..bdb2b0e 100644 (file)
@@ -50,6 +50,19 @@ config MHI_WWAN_MBIM
           To compile this driver as a module, choose M here: the module will be
           called mhi_wwan_mbim.
 
+config QCOM_BAM_DMUX
+       tristate "Qualcomm BAM-DMUX WWAN network driver"
+       depends on (DMA_ENGINE && PM && QCOM_SMEM_STATE) || COMPILE_TEST
+       help
+         The BAM Data Multiplexer provides access to the network data channels
+         of modems integrated into many older Qualcomm SoCs, e.g. Qualcomm
+         MSM8916 or MSM8974. The connection can be established via QMI/AT from
+         userspace with control ports available through the WWAN subsystem
+         (CONFIG_RPMSG_WWAN_CTRL) or QRTR network sockets (CONFIG_QRTR).
+
+         To compile this driver as a module, choose M here: the module will be
+         called qcom_bam_dmux.
+
 config RPMSG_WWAN_CTRL
        tristate "RPMSG WWAN control driver"
        depends on RPMSG
index fe51fee..e722650 100644 (file)
@@ -10,5 +10,6 @@ obj-$(CONFIG_WWAN_HWSIM) += wwan_hwsim.o
 
 obj-$(CONFIG_MHI_WWAN_CTRL) += mhi_wwan_ctrl.o
 obj-$(CONFIG_MHI_WWAN_MBIM) += mhi_wwan_mbim.o
+obj-$(CONFIG_QCOM_BAM_DMUX) += qcom_bam_dmux.o
 obj-$(CONFIG_RPMSG_WWAN_CTRL) += rpmsg_wwan_ctrl.o
 obj-$(CONFIG_IOSM) += iosm/
diff --git a/drivers/net/wwan/qcom_bam_dmux.c b/drivers/net/wwan/qcom_bam_dmux.c
new file mode 100644 (file)
index 0000000..5dfa2eb
--- /dev/null
@@ -0,0 +1,907 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Qualcomm BAM-DMUX WWAN network driver
+ * Copyright (c) 2020, Stephan Gerhold <stephan@gerhold.net>
+ */
+
+#include <linux/atomic.h>
+#include <linux/bitops.h>
+#include <linux/completion.h>
+#include <linux/dma-mapping.h>
+#include <linux/dmaengine.h>
+#include <linux/if_arp.h>
+#include <linux/interrupt.h>
+#include <linux/mod_devicetable.h>
+#include <linux/module.h>
+#include <linux/netdevice.h>
+#include <linux/platform_device.h>
+#include <linux/pm_runtime.h>
+#include <linux/soc/qcom/smem_state.h>
+#include <linux/spinlock.h>
+#include <linux/wait.h>
+#include <linux/workqueue.h>
+#include <net/pkt_sched.h>
+
+#define BAM_DMUX_BUFFER_SIZE           SZ_2K
+#define BAM_DMUX_HDR_SIZE              sizeof(struct bam_dmux_hdr)
+#define BAM_DMUX_MAX_DATA_SIZE         (BAM_DMUX_BUFFER_SIZE - BAM_DMUX_HDR_SIZE)
+#define BAM_DMUX_NUM_SKB               32
+
+#define BAM_DMUX_HDR_MAGIC             0x33fc
+
+#define BAM_DMUX_AUTOSUSPEND_DELAY     1000
+#define BAM_DMUX_REMOTE_TIMEOUT                msecs_to_jiffies(2000)
+
+enum {
+       BAM_DMUX_CMD_DATA,
+       BAM_DMUX_CMD_OPEN,
+       BAM_DMUX_CMD_CLOSE,
+};
+
+enum {
+       BAM_DMUX_CH_DATA_0,
+       BAM_DMUX_CH_DATA_1,
+       BAM_DMUX_CH_DATA_2,
+       BAM_DMUX_CH_DATA_3,
+       BAM_DMUX_CH_DATA_4,
+       BAM_DMUX_CH_DATA_5,
+       BAM_DMUX_CH_DATA_6,
+       BAM_DMUX_CH_DATA_7,
+       BAM_DMUX_NUM_CH
+};
+
+struct bam_dmux_hdr {
+       u16 magic;
+       u8 signal;
+       u8 cmd;
+       u8 pad;
+       u8 ch;
+       u16 len;
+};
+
+struct bam_dmux_skb_dma {
+       struct bam_dmux *dmux;
+       struct sk_buff *skb;
+       dma_addr_t addr;
+};
+
+struct bam_dmux {
+       struct device *dev;
+
+       int pc_irq;
+       bool pc_state, pc_ack_state;
+       struct qcom_smem_state *pc, *pc_ack;
+       u32 pc_mask, pc_ack_mask;
+       wait_queue_head_t pc_wait;
+       struct completion pc_ack_completion;
+
+       struct dma_chan *rx, *tx;
+       struct bam_dmux_skb_dma rx_skbs[BAM_DMUX_NUM_SKB];
+       struct bam_dmux_skb_dma tx_skbs[BAM_DMUX_NUM_SKB];
+       spinlock_t tx_lock; /* Protect tx_skbs, tx_next_skb */
+       unsigned int tx_next_skb;
+       atomic_long_t tx_deferred_skb;
+       struct work_struct tx_wakeup_work;
+
+       DECLARE_BITMAP(remote_channels, BAM_DMUX_NUM_CH);
+       struct work_struct register_netdev_work;
+       struct net_device *netdevs[BAM_DMUX_NUM_CH];
+};
+
+struct bam_dmux_netdev {
+       struct bam_dmux *dmux;
+       u8 ch;
+};
+
+static void bam_dmux_pc_vote(struct bam_dmux *dmux, bool enable)
+{
+       reinit_completion(&dmux->pc_ack_completion);
+       qcom_smem_state_update_bits(dmux->pc, dmux->pc_mask,
+                                   enable ? dmux->pc_mask : 0);
+}
+
+static void bam_dmux_pc_ack(struct bam_dmux *dmux)
+{
+       qcom_smem_state_update_bits(dmux->pc_ack, dmux->pc_ack_mask,
+                                   dmux->pc_ack_state ? 0 : dmux->pc_ack_mask);
+       dmux->pc_ack_state = !dmux->pc_ack_state;
+}
+
+static bool bam_dmux_skb_dma_map(struct bam_dmux_skb_dma *skb_dma,
+                                enum dma_data_direction dir)
+{
+       struct device *dev = skb_dma->dmux->dev;
+
+       skb_dma->addr = dma_map_single(dev, skb_dma->skb->data, skb_dma->skb->len, dir);
+       if (dma_mapping_error(dev, skb_dma->addr)) {
+               dev_err(dev, "Failed to DMA map buffer\n");
+               skb_dma->addr = 0;
+               return false;
+       }
+
+       return true;
+}
+
+static void bam_dmux_skb_dma_unmap(struct bam_dmux_skb_dma *skb_dma,
+                                  enum dma_data_direction dir)
+{
+       dma_unmap_single(skb_dma->dmux->dev, skb_dma->addr, skb_dma->skb->len, dir);
+       skb_dma->addr = 0;
+}
+
+static void bam_dmux_tx_wake_queues(struct bam_dmux *dmux)
+{
+       int i;
+
+       dev_dbg(dmux->dev, "wake queues\n");
+
+       for (i = 0; i < BAM_DMUX_NUM_CH; ++i) {
+               struct net_device *netdev = dmux->netdevs[i];
+
+               if (netdev && netif_running(netdev))
+                       netif_wake_queue(netdev);
+       }
+}
+
+static void bam_dmux_tx_stop_queues(struct bam_dmux *dmux)
+{
+       int i;
+
+       dev_dbg(dmux->dev, "stop queues\n");
+
+       for (i = 0; i < BAM_DMUX_NUM_CH; ++i) {
+               struct net_device *netdev = dmux->netdevs[i];
+
+               if (netdev)
+                       netif_stop_queue(netdev);
+       }
+}
+
+static void bam_dmux_tx_done(struct bam_dmux_skb_dma *skb_dma)
+{
+       struct bam_dmux *dmux = skb_dma->dmux;
+       unsigned long flags;
+
+       pm_runtime_mark_last_busy(dmux->dev);
+       pm_runtime_put_autosuspend(dmux->dev);
+
+       if (skb_dma->addr)
+               bam_dmux_skb_dma_unmap(skb_dma, DMA_TO_DEVICE);
+
+       spin_lock_irqsave(&dmux->tx_lock, flags);
+       skb_dma->skb = NULL;
+       if (skb_dma == &dmux->tx_skbs[dmux->tx_next_skb % BAM_DMUX_NUM_SKB])
+               bam_dmux_tx_wake_queues(dmux);
+       spin_unlock_irqrestore(&dmux->tx_lock, flags);
+}
+
+static void bam_dmux_tx_callback(void *data)
+{
+       struct bam_dmux_skb_dma *skb_dma = data;
+       struct sk_buff *skb = skb_dma->skb;
+
+       bam_dmux_tx_done(skb_dma);
+       dev_consume_skb_any(skb);
+}
+
+static bool bam_dmux_skb_dma_submit_tx(struct bam_dmux_skb_dma *skb_dma)
+{
+       struct bam_dmux *dmux = skb_dma->dmux;
+       struct dma_async_tx_descriptor *desc;
+
+       desc = dmaengine_prep_slave_single(dmux->tx, skb_dma->addr,
+                                          skb_dma->skb->len, DMA_MEM_TO_DEV,
+                                          DMA_PREP_INTERRUPT);
+       if (!desc) {
+               dev_err(dmux->dev, "Failed to prepare TX DMA buffer\n");
+               return false;
+       }
+
+       desc->callback = bam_dmux_tx_callback;
+       desc->callback_param = skb_dma;
+       desc->cookie = dmaengine_submit(desc);
+       return true;
+}
+
+static struct bam_dmux_skb_dma *
+bam_dmux_tx_queue(struct bam_dmux *dmux, struct sk_buff *skb)
+{
+       struct bam_dmux_skb_dma *skb_dma;
+       unsigned long flags;
+
+       spin_lock_irqsave(&dmux->tx_lock, flags);
+
+       skb_dma = &dmux->tx_skbs[dmux->tx_next_skb % BAM_DMUX_NUM_SKB];
+       if (skb_dma->skb) {
+               bam_dmux_tx_stop_queues(dmux);
+               spin_unlock_irqrestore(&dmux->tx_lock, flags);
+               return NULL;
+       }
+       skb_dma->skb = skb;
+
+       dmux->tx_next_skb++;
+       if (dmux->tx_skbs[dmux->tx_next_skb % BAM_DMUX_NUM_SKB].skb)
+               bam_dmux_tx_stop_queues(dmux);
+
+       spin_unlock_irqrestore(&dmux->tx_lock, flags);
+       return skb_dma;
+}
+
+static int bam_dmux_send_cmd(struct bam_dmux_netdev *bndev, u8 cmd)
+{
+       struct bam_dmux *dmux = bndev->dmux;
+       struct bam_dmux_skb_dma *skb_dma;
+       struct bam_dmux_hdr *hdr;
+       struct sk_buff *skb;
+       int ret;
+
+       skb = alloc_skb(sizeof(*hdr), GFP_KERNEL);
+       if (!skb)
+               return -ENOMEM;
+
+       hdr = skb_put_zero(skb, sizeof(*hdr));
+       hdr->magic = BAM_DMUX_HDR_MAGIC;
+       hdr->cmd = cmd;
+       hdr->ch = bndev->ch;
+
+       skb_dma = bam_dmux_tx_queue(dmux, skb);
+       if (!skb_dma) {
+               ret = -EAGAIN;
+               goto free_skb;
+       }
+
+       ret = pm_runtime_get_sync(dmux->dev);
+       if (ret < 0)
+               goto tx_fail;
+
+       if (!bam_dmux_skb_dma_map(skb_dma, DMA_TO_DEVICE)) {
+               ret = -ENOMEM;
+               goto tx_fail;
+       }
+
+       if (!bam_dmux_skb_dma_submit_tx(skb_dma)) {
+               ret = -EIO;
+               goto tx_fail;
+       }
+
+       dma_async_issue_pending(dmux->tx);
+       return 0;
+
+tx_fail:
+       bam_dmux_tx_done(skb_dma);
+free_skb:
+       dev_kfree_skb(skb);
+       return ret;
+}
+
+static int bam_dmux_netdev_open(struct net_device *netdev)
+{
+       struct bam_dmux_netdev *bndev = netdev_priv(netdev);
+       int ret;
+
+       ret = bam_dmux_send_cmd(bndev, BAM_DMUX_CMD_OPEN);
+       if (ret)
+               return ret;
+
+       netif_start_queue(netdev);
+       return 0;
+}
+
+static int bam_dmux_netdev_stop(struct net_device *netdev)
+{
+       struct bam_dmux_netdev *bndev = netdev_priv(netdev);
+
+       netif_stop_queue(netdev);
+       bam_dmux_send_cmd(bndev, BAM_DMUX_CMD_CLOSE);
+       return 0;
+}
+
+static unsigned int needed_room(unsigned int avail, unsigned int needed)
+{
+       if (avail >= needed)
+               return 0;
+       return needed - avail;
+}
+
+static int bam_dmux_tx_prepare_skb(struct bam_dmux_netdev *bndev,
+                                  struct sk_buff *skb)
+{
+       unsigned int head = needed_room(skb_headroom(skb), BAM_DMUX_HDR_SIZE);
+       unsigned int pad = sizeof(u32) - skb->len % sizeof(u32);
+       unsigned int tail = needed_room(skb_tailroom(skb), pad);
+       struct bam_dmux_hdr *hdr;
+       int ret;
+
+       if (head || tail || skb_cloned(skb)) {
+               ret = pskb_expand_head(skb, head, tail, GFP_ATOMIC);
+               if (ret)
+                       return ret;
+       }
+
+       hdr = skb_push(skb, sizeof(*hdr));
+       hdr->magic = BAM_DMUX_HDR_MAGIC;
+       hdr->signal = 0;
+       hdr->cmd = BAM_DMUX_CMD_DATA;
+       hdr->pad = pad;
+       hdr->ch = bndev->ch;
+       hdr->len = skb->len - sizeof(*hdr);
+       if (pad)
+               skb_put_zero(skb, pad);
+
+       return 0;
+}
+
+static netdev_tx_t bam_dmux_netdev_start_xmit(struct sk_buff *skb,
+                                             struct net_device *netdev)
+{
+       struct bam_dmux_netdev *bndev = netdev_priv(netdev);
+       struct bam_dmux *dmux = bndev->dmux;
+       struct bam_dmux_skb_dma *skb_dma;
+       int active, ret;
+
+       skb_dma = bam_dmux_tx_queue(dmux, skb);
+       if (!skb_dma)
+               return NETDEV_TX_BUSY;
+
+       active = pm_runtime_get(dmux->dev);
+       if (active < 0 && active != -EINPROGRESS)
+               goto drop;
+
+       ret = bam_dmux_tx_prepare_skb(bndev, skb);
+       if (ret)
+               goto drop;
+
+       if (!bam_dmux_skb_dma_map(skb_dma, DMA_TO_DEVICE))
+               goto drop;
+
+       if (active <= 0) {
+               /* Cannot sleep here so mark skb for wakeup handler and return */
+               if (!atomic_long_fetch_or(BIT(skb_dma - dmux->tx_skbs),
+                                         &dmux->tx_deferred_skb))
+                       queue_pm_work(&dmux->tx_wakeup_work);
+               return NETDEV_TX_OK;
+       }
+
+       if (!bam_dmux_skb_dma_submit_tx(skb_dma))
+               goto drop;
+
+       dma_async_issue_pending(dmux->tx);
+       return NETDEV_TX_OK;
+
+drop:
+       bam_dmux_tx_done(skb_dma);
+       dev_kfree_skb_any(skb);
+       return NETDEV_TX_OK;
+}
+
+static void bam_dmux_tx_wakeup_work(struct work_struct *work)
+{
+       struct bam_dmux *dmux = container_of(work, struct bam_dmux, tx_wakeup_work);
+       unsigned long pending;
+       int ret, i;
+
+       ret = pm_runtime_resume_and_get(dmux->dev);
+       if (ret < 0) {
+               dev_err(dmux->dev, "Failed to resume: %d\n", ret);
+               return;
+       }
+
+       pending = atomic_long_xchg(&dmux->tx_deferred_skb, 0);
+       if (!pending)
+               goto out;
+
+       dev_dbg(dmux->dev, "pending skbs after wakeup: %#lx\n", pending);
+       for_each_set_bit(i, &pending, BAM_DMUX_NUM_SKB) {
+               bam_dmux_skb_dma_submit_tx(&dmux->tx_skbs[i]);
+       }
+       dma_async_issue_pending(dmux->tx);
+
+out:
+       pm_runtime_mark_last_busy(dmux->dev);
+       pm_runtime_put_autosuspend(dmux->dev);
+}
+
+static const struct net_device_ops bam_dmux_ops = {
+       .ndo_open       = bam_dmux_netdev_open,
+       .ndo_stop       = bam_dmux_netdev_stop,
+       .ndo_start_xmit = bam_dmux_netdev_start_xmit,
+};
+
+static const struct device_type wwan_type = {
+       .name = "wwan",
+};
+
+static void bam_dmux_netdev_setup(struct net_device *dev)
+{
+       dev->netdev_ops = &bam_dmux_ops;
+
+       dev->type = ARPHRD_RAWIP;
+       SET_NETDEV_DEVTYPE(dev, &wwan_type);
+       dev->flags = IFF_POINTOPOINT | IFF_NOARP;
+
+       dev->mtu = ETH_DATA_LEN;
+       dev->max_mtu = BAM_DMUX_MAX_DATA_SIZE;
+       dev->needed_headroom = sizeof(struct bam_dmux_hdr);
+       dev->needed_tailroom = sizeof(u32); /* word-aligned */
+       dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
+
+       /* This perm addr will be used as interface identifier by IPv6 */
+       dev->addr_assign_type = NET_ADDR_RANDOM;
+       eth_random_addr(dev->perm_addr);
+}
+
+static void bam_dmux_register_netdev_work(struct work_struct *work)
+{
+       struct bam_dmux *dmux = container_of(work, struct bam_dmux, register_netdev_work);
+       struct bam_dmux_netdev *bndev;
+       struct net_device *netdev;
+       int ch, ret;
+
+       for_each_set_bit(ch, dmux->remote_channels, BAM_DMUX_NUM_CH) {
+               if (dmux->netdevs[ch])
+                       continue;
+
+               netdev = alloc_netdev(sizeof(*bndev), "wwan%d", NET_NAME_ENUM,
+                                     bam_dmux_netdev_setup);
+               if (!netdev)
+                       return;
+
+               SET_NETDEV_DEV(netdev, dmux->dev);
+               netdev->dev_port = ch;
+
+               bndev = netdev_priv(netdev);
+               bndev->dmux = dmux;
+               bndev->ch = ch;
+
+               ret = register_netdev(netdev);
+               if (ret) {
+                       dev_err(dmux->dev, "Failed to register netdev for channel %u: %d\n",
+                               ch, ret);
+                       free_netdev(netdev);
+                       return;
+               }
+
+               dmux->netdevs[ch] = netdev;
+       }
+}
+
+static void bam_dmux_rx_callback(void *data);
+
+static bool bam_dmux_skb_dma_submit_rx(struct bam_dmux_skb_dma *skb_dma)
+{
+       struct bam_dmux *dmux = skb_dma->dmux;
+       struct dma_async_tx_descriptor *desc;
+
+       desc = dmaengine_prep_slave_single(dmux->rx, skb_dma->addr,
+                                          skb_dma->skb->len, DMA_DEV_TO_MEM,
+                                          DMA_PREP_INTERRUPT);
+       if (!desc) {
+               dev_err(dmux->dev, "Failed to prepare RX DMA buffer\n");
+               return false;
+       }
+
+       desc->callback = bam_dmux_rx_callback;
+       desc->callback_param = skb_dma;
+       desc->cookie = dmaengine_submit(desc);
+       return true;
+}
+
+static bool bam_dmux_skb_dma_queue_rx(struct bam_dmux_skb_dma *skb_dma, gfp_t gfp)
+{
+       if (!skb_dma->skb) {
+               skb_dma->skb = __netdev_alloc_skb(NULL, BAM_DMUX_BUFFER_SIZE, gfp);
+               if (!skb_dma->skb)
+                       return false;
+               skb_put(skb_dma->skb, BAM_DMUX_BUFFER_SIZE);
+       }
+
+       return bam_dmux_skb_dma_map(skb_dma, DMA_FROM_DEVICE) &&
+              bam_dmux_skb_dma_submit_rx(skb_dma);
+}
+
+static void bam_dmux_cmd_data(struct bam_dmux_skb_dma *skb_dma)
+{
+       struct bam_dmux *dmux = skb_dma->dmux;
+       struct sk_buff *skb = skb_dma->skb;
+       struct bam_dmux_hdr *hdr = (struct bam_dmux_hdr *)skb->data;
+       struct net_device *netdev = dmux->netdevs[hdr->ch];
+
+       if (!netdev || !netif_running(netdev)) {
+               dev_warn(dmux->dev, "Data for inactive channel %u\n", hdr->ch);
+               return;
+       }
+
+       if (hdr->len > BAM_DMUX_MAX_DATA_SIZE) {
+               dev_err(dmux->dev, "Data larger than buffer? (%u > %u)\n",
+                       hdr->len, (u16)BAM_DMUX_MAX_DATA_SIZE);
+               return;
+       }
+
+       skb_dma->skb = NULL; /* Hand over to network stack */
+
+       skb_pull(skb, sizeof(*hdr));
+       skb_trim(skb, hdr->len);
+       skb->dev = netdev;
+
+       /* Only Raw-IP/QMAP is supported by this driver */
+       switch (skb->data[0] & 0xf0) {
+       case 0x40:
+               skb->protocol = htons(ETH_P_IP);
+               break;
+       case 0x60:
+               skb->protocol = htons(ETH_P_IPV6);
+               break;
+       default:
+               skb->protocol = htons(ETH_P_MAP);
+               break;
+       }
+
+       netif_receive_skb(skb);
+}
+
+static void bam_dmux_cmd_open(struct bam_dmux *dmux, struct bam_dmux_hdr *hdr)
+{
+       struct net_device *netdev = dmux->netdevs[hdr->ch];
+
+       dev_dbg(dmux->dev, "open channel: %u\n", hdr->ch);
+
+       if (__test_and_set_bit(hdr->ch, dmux->remote_channels)) {
+               dev_warn(dmux->dev, "Channel already open: %u\n", hdr->ch);
+               return;
+       }
+
+       if (netdev) {
+               netif_device_attach(netdev);
+       } else {
+               /* Cannot sleep here, schedule work to register the netdev */
+               schedule_work(&dmux->register_netdev_work);
+       }
+}
+
+static void bam_dmux_cmd_close(struct bam_dmux *dmux, struct bam_dmux_hdr *hdr)
+{
+       struct net_device *netdev = dmux->netdevs[hdr->ch];
+
+       dev_dbg(dmux->dev, "close channel: %u\n", hdr->ch);
+
+       if (!__test_and_clear_bit(hdr->ch, dmux->remote_channels)) {
+               dev_err(dmux->dev, "Channel not open: %u\n", hdr->ch);
+               return;
+       }
+
+       if (netdev)
+               netif_device_detach(netdev);
+}
+
+static void bam_dmux_rx_callback(void *data)
+{
+       struct bam_dmux_skb_dma *skb_dma = data;
+       struct bam_dmux *dmux = skb_dma->dmux;
+       struct sk_buff *skb = skb_dma->skb;
+       struct bam_dmux_hdr *hdr = (struct bam_dmux_hdr *)skb->data;
+
+       bam_dmux_skb_dma_unmap(skb_dma, DMA_FROM_DEVICE);
+
+       if (hdr->magic != BAM_DMUX_HDR_MAGIC) {
+               dev_err(dmux->dev, "Invalid magic in header: %#x\n", hdr->magic);
+               goto out;
+       }
+
+       if (hdr->ch >= BAM_DMUX_NUM_CH) {
+               dev_dbg(dmux->dev, "Unsupported channel: %u\n", hdr->ch);
+               goto out;
+       }
+
+       switch (hdr->cmd) {
+       case BAM_DMUX_CMD_DATA:
+               bam_dmux_cmd_data(skb_dma);
+               break;
+       case BAM_DMUX_CMD_OPEN:
+               bam_dmux_cmd_open(dmux, hdr);
+               break;
+       case BAM_DMUX_CMD_CLOSE:
+               bam_dmux_cmd_close(dmux, hdr);
+               break;
+       default:
+               dev_err(dmux->dev, "Unsupported command %u on channel %u\n",
+                       hdr->cmd, hdr->ch);
+               break;
+       }
+
+out:
+       if (bam_dmux_skb_dma_queue_rx(skb_dma, GFP_ATOMIC))
+               dma_async_issue_pending(dmux->rx);
+}
+
+static bool bam_dmux_power_on(struct bam_dmux *dmux)
+{
+       struct device *dev = dmux->dev;
+       struct dma_slave_config dma_rx_conf = {
+               .direction = DMA_DEV_TO_MEM,
+               .src_maxburst = BAM_DMUX_BUFFER_SIZE,
+       };
+       int i;
+
+       dmux->rx = dma_request_chan(dev, "rx");
+       if (IS_ERR(dmux->rx)) {
+               dev_err(dev, "Failed to request RX DMA channel: %pe\n", dmux->rx);
+               dmux->rx = NULL;
+               return false;
+       }
+       dmaengine_slave_config(dmux->rx, &dma_rx_conf);
+
+       for (i = 0; i < BAM_DMUX_NUM_SKB; i++) {
+               if (!bam_dmux_skb_dma_queue_rx(&dmux->rx_skbs[i], GFP_KERNEL))
+                       return false;
+       }
+       dma_async_issue_pending(dmux->rx);
+
+       return true;
+}
+
+static void bam_dmux_free_skbs(struct bam_dmux_skb_dma skbs[],
+                              enum dma_data_direction dir)
+{
+       int i;
+
+       for (i = 0; i < BAM_DMUX_NUM_SKB; i++) {
+               struct bam_dmux_skb_dma *skb_dma = &skbs[i];
+
+               if (skb_dma->addr)
+                       bam_dmux_skb_dma_unmap(skb_dma, dir);
+               if (skb_dma->skb) {
+                       dev_kfree_skb(skb_dma->skb);
+                       skb_dma->skb = NULL;
+               }
+       }
+}
+
+static void bam_dmux_power_off(struct bam_dmux *dmux)
+{
+       if (dmux->tx) {
+               dmaengine_terminate_sync(dmux->tx);
+               dma_release_channel(dmux->tx);
+               dmux->tx = NULL;
+       }
+
+       if (dmux->rx) {
+               dmaengine_terminate_sync(dmux->rx);
+               dma_release_channel(dmux->rx);
+               dmux->rx = NULL;
+       }
+
+       bam_dmux_free_skbs(dmux->rx_skbs, DMA_FROM_DEVICE);
+}
+
+static irqreturn_t bam_dmux_pc_irq(int irq, void *data)
+{
+       struct bam_dmux *dmux = data;
+       bool new_state = !dmux->pc_state;
+
+       dev_dbg(dmux->dev, "pc: %u\n", new_state);
+
+       if (new_state) {
+               if (bam_dmux_power_on(dmux))
+                       bam_dmux_pc_ack(dmux);
+               else
+                       bam_dmux_power_off(dmux);
+       } else {
+               bam_dmux_power_off(dmux);
+               bam_dmux_pc_ack(dmux);
+       }
+
+       dmux->pc_state = new_state;
+       wake_up_all(&dmux->pc_wait);
+
+       return IRQ_HANDLED;
+}
+
+static irqreturn_t bam_dmux_pc_ack_irq(int irq, void *data)
+{
+       struct bam_dmux *dmux = data;
+
+       dev_dbg(dmux->dev, "pc ack\n");
+       complete_all(&dmux->pc_ack_completion);
+
+       return IRQ_HANDLED;
+}
+
+static int bam_dmux_runtime_suspend(struct device *dev)
+{
+       struct bam_dmux *dmux = dev_get_drvdata(dev);
+
+       dev_dbg(dev, "runtime suspend\n");
+       bam_dmux_pc_vote(dmux, false);
+
+       return 0;
+}
+
+static int __maybe_unused bam_dmux_runtime_resume(struct device *dev)
+{
+       struct bam_dmux *dmux = dev_get_drvdata(dev);
+
+       dev_dbg(dev, "runtime resume\n");
+
+       /* Wait until previous power down was acked */
+       if (!wait_for_completion_timeout(&dmux->pc_ack_completion,
+                                        BAM_DMUX_REMOTE_TIMEOUT))
+               return -ETIMEDOUT;
+
+       /* Vote for power state */
+       bam_dmux_pc_vote(dmux, true);
+
+       /* Wait for ack */
+       if (!wait_for_completion_timeout(&dmux->pc_ack_completion,
+                                        BAM_DMUX_REMOTE_TIMEOUT)) {
+               bam_dmux_pc_vote(dmux, false);
+               return -ETIMEDOUT;
+       }
+
+       /* Wait until we're up */
+       if (!wait_event_timeout(dmux->pc_wait, dmux->pc_state,
+                               BAM_DMUX_REMOTE_TIMEOUT)) {
+               bam_dmux_pc_vote(dmux, false);
+               return -ETIMEDOUT;
+       }
+
+       /* Ensure that we actually initialized successfully */
+       if (!dmux->rx) {
+               bam_dmux_pc_vote(dmux, false);
+               return -ENXIO;
+       }
+
+       /* Request TX channel if necessary */
+       if (dmux->tx)
+               return 0;
+
+       dmux->tx = dma_request_chan(dev, "tx");
+       if (IS_ERR(dmux->rx)) {
+               dev_err(dev, "Failed to request TX DMA channel: %pe\n", dmux->tx);
+               dmux->tx = NULL;
+               bam_dmux_runtime_suspend(dev);
+               return -ENXIO;
+       }
+
+       return 0;
+}
+
+static int bam_dmux_probe(struct platform_device *pdev)
+{
+       struct device *dev = &pdev->dev;
+       struct bam_dmux *dmux;
+       int ret, pc_ack_irq, i;
+       unsigned int bit;
+
+       dmux = devm_kzalloc(dev, sizeof(*dmux), GFP_KERNEL);
+       if (!dmux)
+               return -ENOMEM;
+
+       dmux->dev = dev;
+       platform_set_drvdata(pdev, dmux);
+
+       dmux->pc_irq = platform_get_irq_byname(pdev, "pc");
+       if (dmux->pc_irq < 0)
+               return dmux->pc_irq;
+
+       pc_ack_irq = platform_get_irq_byname(pdev, "pc-ack");
+       if (pc_ack_irq < 0)
+               return pc_ack_irq;
+
+       dmux->pc = devm_qcom_smem_state_get(dev, "pc", &bit);
+       if (IS_ERR(dmux->pc))
+               return dev_err_probe(dev, PTR_ERR(dmux->pc),
+                                    "Failed to get pc state\n");
+       dmux->pc_mask = BIT(bit);
+
+       dmux->pc_ack = devm_qcom_smem_state_get(dev, "pc-ack", &bit);
+       if (IS_ERR(dmux->pc_ack))
+               return dev_err_probe(dev, PTR_ERR(dmux->pc_ack),
+                                    "Failed to get pc-ack state\n");
+       dmux->pc_ack_mask = BIT(bit);
+
+       init_waitqueue_head(&dmux->pc_wait);
+       init_completion(&dmux->pc_ack_completion);
+       complete_all(&dmux->pc_ack_completion);
+
+       spin_lock_init(&dmux->tx_lock);
+       INIT_WORK(&dmux->tx_wakeup_work, bam_dmux_tx_wakeup_work);
+       INIT_WORK(&dmux->register_netdev_work, bam_dmux_register_netdev_work);
+
+       for (i = 0; i < BAM_DMUX_NUM_SKB; i++) {
+               dmux->rx_skbs[i].dmux = dmux;
+               dmux->tx_skbs[i].dmux = dmux;
+       }
+
+       /* Runtime PM manages our own power vote.
+        * Note that the RX path may be active even if we are runtime suspended,
+        * since it is controlled by the remote side.
+        */
+       pm_runtime_set_autosuspend_delay(dev, BAM_DMUX_AUTOSUSPEND_DELAY);
+       pm_runtime_use_autosuspend(dev);
+       pm_runtime_enable(dev);
+
+       ret = devm_request_threaded_irq(dev, pc_ack_irq, NULL, bam_dmux_pc_ack_irq,
+                                       IRQF_ONESHOT, NULL, dmux);
+       if (ret)
+               return ret;
+
+       ret = devm_request_threaded_irq(dev, dmux->pc_irq, NULL, bam_dmux_pc_irq,
+                                       IRQF_ONESHOT, NULL, dmux);
+       if (ret)
+               return ret;
+
+       ret = irq_get_irqchip_state(dmux->pc_irq, IRQCHIP_STATE_LINE_LEVEL,
+                                   &dmux->pc_state);
+       if (ret)
+               return ret;
+
+       /* Check if remote finished initialization before us */
+       if (dmux->pc_state) {
+               if (bam_dmux_power_on(dmux))
+                       bam_dmux_pc_ack(dmux);
+               else
+                       bam_dmux_power_off(dmux);
+       }
+
+       return 0;
+}
+
+static int bam_dmux_remove(struct platform_device *pdev)
+{
+       struct bam_dmux *dmux = platform_get_drvdata(pdev);
+       struct device *dev = dmux->dev;
+       LIST_HEAD(list);
+       int i;
+
+       /* Unregister network interfaces */
+       cancel_work_sync(&dmux->register_netdev_work);
+       rtnl_lock();
+       for (i = 0; i < BAM_DMUX_NUM_CH; ++i)
+               if (dmux->netdevs[i])
+                       unregister_netdevice_queue(dmux->netdevs[i], &list);
+       unregister_netdevice_many(&list);
+       rtnl_unlock();
+       cancel_work_sync(&dmux->tx_wakeup_work);
+
+       /* Drop our own power vote */
+       pm_runtime_disable(dev);
+       pm_runtime_dont_use_autosuspend(dev);
+       bam_dmux_runtime_suspend(dev);
+       pm_runtime_set_suspended(dev);
+
+       /* Try to wait for remote side to drop power vote */
+       if (!wait_event_timeout(dmux->pc_wait, !dmux->rx, BAM_DMUX_REMOTE_TIMEOUT))
+               dev_err(dev, "Timed out waiting for remote side to suspend\n");
+
+       /* Make sure everything is cleaned up before we return */
+       disable_irq(dmux->pc_irq);
+       bam_dmux_power_off(dmux);
+       bam_dmux_free_skbs(dmux->tx_skbs, DMA_TO_DEVICE);
+
+       return 0;
+}
+
+static const struct dev_pm_ops bam_dmux_pm_ops = {
+       SET_RUNTIME_PM_OPS(bam_dmux_runtime_suspend, bam_dmux_runtime_resume, NULL)
+};
+
+static const struct of_device_id bam_dmux_of_match[] = {
+       { .compatible = "qcom,bam-dmux" },
+       { /* sentinel */ }
+};
+MODULE_DEVICE_TABLE(of, bam_dmux_of_match);
+
+static struct platform_driver bam_dmux_driver = {
+       .probe = bam_dmux_probe,
+       .remove = bam_dmux_remove,
+       .driver = {
+               .name = "bam-dmux",
+               .pm = &bam_dmux_pm_ops,
+               .of_match_table = bam_dmux_of_match,
+       },
+};
+module_platform_driver(bam_dmux_driver);
+
+MODULE_LICENSE("GPL v2");
+MODULE_DESCRIPTION("Qualcomm BAM-DMUX WWAN Network Driver");
+MODULE_AUTHOR("Stephan Gerhold <stephan@gerhold.net>");