/* Copyright (c) 2014-2015, The Linux Foundation. All rights reserved. * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 and * only version 2 as published by the Free Software Foundation. * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "glink_core_if.h" #include "glink_private.h" #include "glink_xprt_if.h" #define XPRT_NAME "smem" #define FIFO_FULL_RESERVE 8 #define FIFO_ALIGNMENT 8 #define TX_BLOCKED_CMD_RESERVE 8 /* size of struct read_notif_request */ #define SMEM_CH_DESC_SIZE 32 #define RPM_TOC_ID 0x67727430 #define RPM_TX_FIFO_ID 0x61703272 #define RPM_RX_FIFO_ID 0x72326170 #define RPM_TOC_SIZE 256 #define RPM_MAX_TOC_ENTRIES 20 #define RPM_FIFO_ADDR_ALIGN_BYTES 3 #define TRACER_PKT_FEATURE BIT(2) #define NUM_SMEM_SUBSYSTEMS 3 /** * enum command_types - definition of the types of commands sent/received * @VERSION_CMD: Version and feature set supported * @VERSION_ACK_CMD: Response for @VERSION_CMD * @OPEN_CMD: Open a channel * @CLOSE_CMD: Close a channel * @OPEN_ACK_CMD: Response to @OPEN_CMD * @RX_INTENT_CMD: RX intent for a channel was queued * @RX_DONE_CMD: Use of RX intent for a channel is complete * @RX_INTENT_REQ_CMD: Request to have RX intent queued * @RX_INTENT_REQ_ACK_CMD: Response for @RX_INTENT_REQ_CMD * @TX_DATA_CMD: Start of a data transfer * @ZERO_COPY_TX_DATA_CMD: Start of a data transfer with zero copy * @CLOSE_ACK_CMD: Response for @CLOSE_CMD * @TX_DATA_CONT_CMD: Continuation or end of a data transfer * @READ_NOTIF_CMD: Request for a notification when this cmd is read * @RX_DONE_W_REUSE_CMD: Same as @RX_DONE but also reuse the used intent * @SIGNALS_CMD: Sideband signals * @TRACER_PKT_CMD: Start of a Tracer Packet Command * @TRACER_PKT_CONT_CMD: Continuation or end of a Tracer Packet Command */ enum command_types { VERSION_CMD, VERSION_ACK_CMD, OPEN_CMD, CLOSE_CMD, OPEN_ACK_CMD, RX_INTENT_CMD, RX_DONE_CMD, RX_INTENT_REQ_CMD, RX_INTENT_REQ_ACK_CMD, TX_DATA_CMD, ZERO_COPY_TX_DATA_CMD, CLOSE_ACK_CMD, TX_DATA_CONT_CMD, READ_NOTIF_CMD, RX_DONE_W_REUSE_CMD, SIGNALS_CMD, TRACER_PKT_CMD, TRACER_PKT_CONT_CMD, }; /** * struct channel_desc - description of a channel fifo with a remote entity * @read_index: The read index for the fifo where data should be * consumed from. * @write_index: The write index for the fifo where data should produced * to. * * This structure resides in SMEM and contains the control information for the * fifo data pipes of the channel. There is one physical channel between us * and a remote entity. */ struct channel_desc { uint32_t read_index; uint32_t write_index; }; /** * struct mailbox_config_info - description of a mailbox tranposrt channel * @tx_read_index: Offset into the tx fifo where data should be read from. * @tx_write_index: Offset into the tx fifo where new data will be placed. * @tx_size: Size of the transmit fifo in bytes. * @rx_read_index: Offset into the rx fifo where data should be read from. * @rx_write_index: Offset into the rx fifo where new data will be placed. * @rx_size: Size of the receive fifo in bytes. * @fifo: The fifos for the channel. */ struct mailbox_config_info { uint32_t tx_read_index; uint32_t tx_write_index; uint32_t tx_size; uint32_t rx_read_index; uint32_t rx_write_index; uint32_t rx_size; char fifo[]; /* tx fifo, then rx fifo */ }; /** * struct edge_info - local information for managing a single complete edge * @xprt_if: The transport interface registered with the * glink core associated with this edge. * @xprt_cfg: The transport configuration for the glink core * assocaited with this edge. * @intentless: True if this edge runs in intentless mode. * @irq_disabled: Flag indicating the whether interrupt is enabled * or disabled. * @remote_proc_id: The SMEM processor id for the remote side. * @rx_reset_reg: Reference to the register to reset the rx irq * line, if applicable. * @out_irq_reg: Reference to the register to send an irq to the * remote side. * @out_irq_mask: Mask written to @out_irq_reg to trigger the * correct irq. * @irq_line: The incoming interrupt line. * @tx_irq_count: Number of interrupts triggered. * @rx_irq_count: Number of interrupts received. * @tx_ch_desc: Reference to the channel description structure * for tx in SMEM for this edge. * @rx_ch_desc: Reference to the channel description structure * for rx in SMEM for this edge. * @tx_fifo: Reference to the transmit fifo in SMEM. * @rx_fifo: Reference to the receive fifo in SMEM. * @tx_fifo_size: Total size of @tx_fifo. * @rx_fifo_size: Total size of @rx_fifo. * @read_from_fifo: Memcpy for this edge. * @write_to_fifo: Memcpy for this edge. * @write_lock: Lock to serialize access to @tx_fifo. * @tx_blocked_queue: Queue of entities waiting for the remote side to * signal @tx_fifo has flushed and is now empty. * @tx_resume_needed: A tx resume signal needs to be sent to the glink * core once the remote side indicates @tx_fifo has * flushed. * @tx_blocked_signal_sent: Flag to indicate the flush signal has already * been sent, and a response is pending from the * remote side. Protected by @write_lock. * @kwork: Work to be executed when an irq is received. * @kworker: Handle to the entity processing @kwork. * @task: Handle to the task context used to run @kworker. * @use_ref: Active uses of this transport use this to grab * a reference. Used for ssr synchronization. * @in_ssr: Signals if this transport is in ssr. * @rx_lock: Used to serialize concurrent instances of rx * processing. * @deferred_cmds: List of deferred commands that need to be * processed in process context. * @num_pw_states: Size of @ramp_time_us. * @ramp_time_us: Array of ramp times in microseconds where array * index position represents a power state. * @mailbox: Mailbox transport channel description reference. */ struct edge_info { struct glink_transport_if xprt_if; struct glink_core_transport_cfg xprt_cfg; bool intentless; bool irq_disabled; uint32_t remote_proc_id; void __iomem *rx_reset_reg; void __iomem *out_irq_reg; uint32_t out_irq_mask; uint32_t irq_line; uint32_t tx_irq_count; uint32_t rx_irq_count; struct channel_desc *tx_ch_desc; struct channel_desc *rx_ch_desc; void __iomem *tx_fifo; void __iomem *rx_fifo; size_t tx_fifo_size; size_t rx_fifo_size; void * (*read_from_fifo)(void *dest, const void *src, size_t num_bytes); void * (*write_to_fifo)(void *dest, const void *src, size_t num_bytes); spinlock_t write_lock; wait_queue_head_t tx_blocked_queue; bool tx_resume_needed; bool tx_blocked_signal_sent; struct kthread_work kwork; struct kthread_worker kworker; struct task_struct *task; struct srcu_struct use_ref; bool in_ssr; spinlock_t rx_lock; struct list_head deferred_cmds; uint32_t num_pw_states; unsigned long *ramp_time_us; struct mailbox_config_info *mailbox; uint32_t tx_fifo_number; uint32_t rx_fifo_number; }; /** * struct deferred_cmd - description of a command to be processed later * @list_node: Used to put this command on a list in the edge. * @id: ID of the command. * @param1: Parameter one of the command. * @param2: Parameter two of the command. * @data: Extra data associated with the command, if applicable. * * This structure stores the relevant information of a command that was removed * from the fifo but needs to be processed at a later time. */ struct deferred_cmd { struct list_head list_node; uint16_t id; uint16_t param1; uint32_t param2; void *data; }; static uint32_t negotiate_features_v1(struct glink_transport_if *if_ptr, const struct glink_core_version *version, uint32_t features); static void register_debugfs_info(struct edge_info *einfo); static int ssr(struct glink_transport_if *if_ptr); static struct edge_info *edge_infos[NUM_SMEM_SUBSYSTEMS]; static DEFINE_MUTEX(probe_lock); static struct glink_core_version versions[] = { {1, TRACER_PKT_FEATURE, negotiate_features_v1}, }; static struct platform_driver glink_smem_native_driver; /** * send_irq() - send an irq to a remote entity as an event signal * @einfo: Which remote entity that should receive the irq. */ static void send_irq(struct edge_info *einfo) { /* * Any data associated with this event must be visable to the remote * before the interrupt is triggered */ wmb(); writel_relaxed(einfo->out_irq_mask, einfo->out_irq_reg); einfo->tx_irq_count++; } /** * read_from_fifo() - memcpy from fifo memory * @dest: Destination address. * @src: Source address. * @num_bytes: Number of bytes to copy. * * Return: Destination address. */ static void *read_from_fifo(void *dest, const void *src, size_t num_bytes) { memcpy_fromio(dest, src, num_bytes); return dest; } /** * write_to_fifo() - memcpy to fifo memory * @dest: Destination address. * @src: Source address. * @num_bytes: Number of bytes to copy. * * Return: Destination address. */ static void *write_to_fifo(void *dest, const void *src, size_t num_bytes) { memcpy_toio(dest, src, num_bytes); return dest; } /** * memcpy32_toio() - memcpy to word access only memory * @dest: Destination address. * @src: Source address. * @num_bytes: Number of bytes to copy. * * Return: Destination address. */ static void *memcpy32_toio(void *dest, const void *src, size_t num_bytes) { uint32_t *dest_local = (uint32_t *)dest; uint32_t *src_local = (uint32_t *)src; BUG_ON(num_bytes & RPM_FIFO_ADDR_ALIGN_BYTES); BUG_ON(!dest_local || ((uintptr_t)dest_local & RPM_FIFO_ADDR_ALIGN_BYTES)); BUG_ON(!src_local || ((uintptr_t)src_local & RPM_FIFO_ADDR_ALIGN_BYTES)); num_bytes /= sizeof(uint32_t); while (num_bytes--) __raw_writel(*src_local++, dest_local++); return dest; } /** * memcpy32_fromio() - memcpy from word access only memory * @dest: Destination address. * @src: Source address. * @num_bytes: Number of bytes to copy. * * Return: Destination address. */ static void *memcpy32_fromio(void *dest, const void *src, size_t num_bytes) { uint32_t *dest_local = (uint32_t *)dest; uint32_t *src_local = (uint32_t *)src; BUG_ON(num_bytes & RPM_FIFO_ADDR_ALIGN_BYTES); BUG_ON(!dest_local || ((uintptr_t)dest_local & RPM_FIFO_ADDR_ALIGN_BYTES)); BUG_ON(!src_local || ((uintptr_t)src_local & RPM_FIFO_ADDR_ALIGN_BYTES)); num_bytes /= sizeof(uint32_t); while (num_bytes--) *dest_local++ = __raw_readl(src_local++); return dest; } /** * fifo_read_avail() - how many bytes are available to be read from an edge * @einfo: The concerned edge to query. * * Return: The number of bytes available to be read from edge. */ static uint32_t fifo_read_avail(struct edge_info *einfo) { uint32_t read_index = einfo->rx_ch_desc->read_index; uint32_t write_index = einfo->rx_ch_desc->write_index; uint32_t fifo_size = einfo->rx_fifo_size; uint32_t bytes_avail; bytes_avail = write_index - read_index; if (write_index < read_index) /* * Case: W < R - Write has wrapped * -------------------------------- * In this case, the write operation has wrapped past the end * of the FIFO which means that now calculating the amount of * data in the FIFO results in a negative number. This can be * easily fixed by adding the fifo_size to the value. Even * though the values are unsigned, subtraction is always done * using 2's complement which means that the result will still * be correct once the FIFO size has been added to the negative * result. * * Example: * '-' = data in fifo * '.' = empty * * 0 1 * 0123456789012345 * |-----w.....r----| * 0 N * * write = 5 = 101b * read = 11 = 1011b * Data in FIFO * (write - read) + fifo_size = (101b - 1011b) + 10000b * = 11111010b + 10000b = 1010b = 10 */ bytes_avail += fifo_size; return bytes_avail; } /** * fifo_write_avail() - how many bytes can be written to the edge * @einfo: The concerned edge to query. * * Calculates the number of bytes that can be transmitted at this time. * Automatically reserves some space to maintain alignment when the fifo is * completely full, and reserves space so that the flush command can always be * transmitted when needed. * * Return: The number of bytes available to be read from edge. */ static uint32_t fifo_write_avail(struct edge_info *einfo) { uint32_t read_index = einfo->tx_ch_desc->read_index; uint32_t write_index = einfo->tx_ch_desc->write_index; uint32_t fifo_size = einfo->tx_fifo_size; uint32_t bytes_avail = read_index - write_index; if (read_index <= write_index) bytes_avail += fifo_size; if (bytes_avail < FIFO_FULL_RESERVE + TX_BLOCKED_CMD_RESERVE) bytes_avail = 0; else bytes_avail -= FIFO_FULL_RESERVE + TX_BLOCKED_CMD_RESERVE; return bytes_avail; } /** * fifo_read() - read data from an edge * @einfo: The concerned edge to read from. * @_data: Buffer to copy the read data into. * @len: The ammount of data to read in bytes. * * Return: The number of bytes read. */ static int fifo_read(struct edge_info *einfo, void *_data, int len) { void *ptr; void *data = _data; int orig_len = len; uint32_t read_index = einfo->rx_ch_desc->read_index; uint32_t write_index = einfo->rx_ch_desc->write_index; uint32_t fifo_size = einfo->rx_fifo_size; uint32_t n; while (len) { ptr = einfo->rx_fifo + read_index; if (read_index <= write_index) n = write_index - read_index; else n = fifo_size - read_index; if (n == 0) break; if (n > len) n = len; einfo->read_from_fifo(data, ptr, n); data += n; len -= n; read_index += n; if (read_index >= fifo_size) read_index -= fifo_size; } einfo->rx_ch_desc->read_index = read_index; return orig_len - len; } /** * fifo_write_body() - Copy transmit data into an edge * @einfo: The concerned edge to copy into. * @_data: Buffer of data to copy from. * @len: Size of data to copy in bytes. * @write_index: Index into the channel where the data should be copied. * * Return: Number of bytes remaining to be copied into the edge. */ static uint32_t fifo_write_body(struct edge_info *einfo, const void *_data, int len, uint32_t *write_index) { void *ptr; const void *data = _data; uint32_t read_index = einfo->tx_ch_desc->read_index; uint32_t fifo_size = einfo->tx_fifo_size; uint32_t n; while (len) { ptr = einfo->tx_fifo + *write_index; if (*write_index < read_index) { n = read_index - *write_index - FIFO_FULL_RESERVE; } else { if (read_index < FIFO_FULL_RESERVE) n = fifo_size + read_index - *write_index - FIFO_FULL_RESERVE; else n = fifo_size - *write_index; } if (n == 0) break; if (n > len) n = len; einfo->write_to_fifo(ptr, data, n); data += n; len -= n; *write_index += n; if (*write_index >= fifo_size) *write_index -= fifo_size; } return len; } /** * fifo_write() - Write data into an edge * @einfo: The concerned edge to write to. * @data: Buffer of data to write. * @len: Length of data to write, in bytes. * * Wrapper around fifo_write_body() to manage additional details that are * necessary for a complete write event. Does not manage concurrency. Clients * should use fifo_write_avail() to check if there is sufficent space before * calling fifo_write(). * * Return: Number of bytes written to the edge. */ static int fifo_write(struct edge_info *einfo, const void *data, int len) { int orig_len = len; uint32_t write_index = einfo->tx_ch_desc->write_index; len = fifo_write_body(einfo, data, len, &write_index); einfo->tx_ch_desc->write_index = write_index; send_irq(einfo); return orig_len - len; } /** * fifo_write_complex() - writes a transaction of multiple buffers to an edge * @einfo: The concerned edge to write to. * @data1: The first buffer of data to write. * @len1: The length of the first buffer in bytes. * @data2: The second buffer of data to write. * @len2: The length of the second buffer in bytes. * @data3: The thirs buffer of data to write. * @len3: The length of the third buffer in bytes. * * A variant of fifo_write() which optimizes the usecase found in tx(). The * remote side expects all or none of the transmitted data to be available. * This prevents the tx() usecase from calling fifo_write() multiple times. The * alternative would be an allocation and additional memcpy to create a buffer * to copy all the data segments into one location before calling fifo_write(). * * Return: Number of bytes written to the edge. */ static int fifo_write_complex(struct edge_info *einfo, const void *data1, int len1, const void *data2, int len2, const void *data3, int len3) { int orig_len = len1 + len2 + len3; uint32_t write_index = einfo->tx_ch_desc->write_index; len1 = fifo_write_body(einfo, data1, len1, &write_index); len2 = fifo_write_body(einfo, data2, len2, &write_index); len3 = fifo_write_body(einfo, data3, len3, &write_index); einfo->tx_ch_desc->write_index = write_index; send_irq(einfo); return orig_len - len1 - len2 - len3; } /** * send_tx_blocked_signal() - send the flush command as we are blocked from tx * @einfo: The concerned edge which is blocked. * * Used to send a signal to the remote side that we have no more space to * transmit data and therefore need the remote side to signal us when they have * cleared some space by reading some data. This function relies upon the * assumption that fifo_write_avail() will reserve some space so that the flush * signal command can always be put into the transmit fifo, even when "everyone" * else thinks that the transmit fifo is truely full. This function assumes * that it is called with the write_lock already locked. */ static void send_tx_blocked_signal(struct edge_info *einfo) { struct read_notif_request { uint16_t cmd; uint16_t reserved; uint32_t reserved2; }; struct read_notif_request read_notif_req; read_notif_req.cmd = READ_NOTIF_CMD; read_notif_req.reserved = 0; read_notif_req.reserved2 = 0; if (!einfo->tx_blocked_signal_sent) { einfo->tx_blocked_signal_sent = true; fifo_write(einfo, &read_notif_req, sizeof(read_notif_req)); } } /** * fifo_tx() - transmit data on an edge * @einfo: The concerned edge to transmit on. * @data: Buffer of data to transmit. * @len: Length of data to transmit in bytes. * * This helper function is the preferred interface to fifo_write() and should * be used in the normal case for transmitting entities. fifo_tx() will block * until there is sufficent room to transmit the requested ammount of data. * fifo_tx() will manage any concurrency between multiple transmitters on a * channel. * * Return: Number of bytes transmitted. */ static int fifo_tx(struct edge_info *einfo, const void *data, int len) { unsigned long flags; int ret; DEFINE_WAIT(wait); spin_lock_irqsave(&einfo->write_lock, flags); while (fifo_write_avail(einfo) < len) { send_tx_blocked_signal(einfo); spin_unlock_irqrestore(&einfo->write_lock, flags); prepare_to_wait(&einfo->tx_blocked_queue, &wait, TASK_UNINTERRUPTIBLE); if (fifo_write_avail(einfo) < len && !einfo->in_ssr) schedule(); finish_wait(&einfo->tx_blocked_queue, &wait); spin_lock_irqsave(&einfo->write_lock, flags); if (einfo->in_ssr) { spin_unlock_irqrestore(&einfo->write_lock, flags); return -EFAULT; } } ret = fifo_write(einfo, data, len); spin_unlock_irqrestore(&einfo->write_lock, flags); return ret; } /** * process_rx_data() - process received data from an edge * @einfo: The edge the data was received on. * @cmd_id: ID to specify the type of data. * @rcid: The remote channel id associated with the data. * @intend_id: The intent the data should be put in. */ static void process_rx_data(struct edge_info *einfo, uint16_t cmd_id, uint32_t rcid, uint32_t intent_id) { struct command { uint32_t frag_size; uint32_t size_remaining; }; struct command cmd; struct glink_core_rx_intent *intent; char trash[FIFO_ALIGNMENT]; int alignment; bool err = false; fifo_read(einfo, &cmd, sizeof(cmd)); intent = einfo->xprt_if.glink_core_if_ptr->rx_get_pkt_ctx( &einfo->xprt_if, rcid, intent_id); if (intent == NULL) { GLINK_ERR("%s: no intent for ch %d liid %d\n", __func__, rcid, intent_id); err = true; } else if (intent->data == NULL) { if (einfo->intentless) { intent->data = kmalloc(cmd.frag_size, GFP_ATOMIC); if (!intent->data) err = true; else intent->intent_size = cmd.frag_size; } else { GLINK_ERR( "%s: intent for ch %d liid %d has no data buff\n", __func__, rcid, intent_id); err = true; } } if (!err && (intent->intent_size - intent->write_offset < cmd.frag_size || intent->write_offset + cmd.size_remaining > intent->intent_size)) { GLINK_ERR("%s: rx data size:%d and remaining:%d %s %d %s:%d\n", __func__, cmd.frag_size, cmd.size_remaining, "will overflow ch", rcid, "intent", intent_id); err = true; } if (err) { alignment = ALIGN(cmd.frag_size, FIFO_ALIGNMENT); alignment -= cmd.frag_size; while (cmd.frag_size) { if (cmd.frag_size > FIFO_ALIGNMENT) { fifo_read(einfo, trash, FIFO_ALIGNMENT); cmd.frag_size -= FIFO_ALIGNMENT; } else { fifo_read(einfo, trash, cmd.frag_size); cmd.frag_size = 0; } } if (alignment) fifo_read(einfo, trash, alignment); return; } fifo_read(einfo, intent->data + intent->write_offset, cmd.frag_size); intent->write_offset += cmd.frag_size; intent->pkt_size += cmd.frag_size; alignment = ALIGN(cmd.frag_size, FIFO_ALIGNMENT); alignment -= cmd.frag_size; if (alignment) fifo_read(einfo, trash, alignment); if (unlikely((cmd_id == TRACER_PKT_CMD || cmd_id == TRACER_PKT_CONT_CMD) && !cmd.size_remaining)) { tracer_pkt_log_event(intent->data, GLINK_XPRT_RX); intent->tracer_pkt = true; } einfo->xprt_if.glink_core_if_ptr->rx_put_pkt_ctx(&einfo->xprt_if, rcid, intent, cmd.size_remaining ? false : true); } /** * queue_cmd() - queue a deferred command for later processing * @einfo: Edge to queue commands on. * @cmd: Command to queue. * @data: Command specific data to queue with the command. * * Return: True if queuing was successful, false otherwise. */ static bool queue_cmd(struct edge_info *einfo, void *cmd, void *data) { struct command { uint16_t id; uint16_t param1; uint32_t param2; }; struct command *_cmd = cmd; struct deferred_cmd *d_cmd; d_cmd = kmalloc(sizeof(*d_cmd), GFP_ATOMIC); if (!d_cmd) { GLINK_ERR("%s: Discarding cmd %d\n", __func__, _cmd->id); return false; } d_cmd->id = _cmd->id; d_cmd->param1 = _cmd->param1; d_cmd->param2 = _cmd->param2; d_cmd->data = data; list_add_tail(&d_cmd->list_node, &einfo->deferred_cmds); queue_kthread_work(&einfo->kworker, &einfo->kwork); return true; } /** * get_rx_fifo() - Find the rx fifo for an edge * @einfo: Edge to find the fifo for. * * Return: True if fifo was found, false otherwise. */ static bool get_rx_fifo(struct edge_info *einfo) { void __iomem *rx_fifo = qcom_smem_get(einfo->remote_proc_id, einfo->rx_fifo_number, &einfo->rx_fifo_size); if (!rx_fifo || IS_ERR(rx_fifo)) return false; einfo->rx_fifo = rx_fifo; return true; } /** * __rx_worker() - process received commands on a specific edge * @einfo: Edge to process commands on. * @atomic_ctx: Indicates if the caller is in atomic context and requires any * non-atomic operations to be deferred. */ static void __rx_worker(struct edge_info *einfo, bool atomic_ctx) { struct command { uint16_t id; uint16_t param1; uint32_t param2; }; struct intent_desc { uint32_t size; uint32_t id; }; struct command cmd; struct intent_desc intent; struct intent_desc *intents; int i; bool granted; unsigned long flags; bool trigger_wakeup = false; int rcu_id; uint16_t rcid; uint32_t name_len; uint32_t len; char *name; char trash[FIFO_ALIGNMENT]; struct deferred_cmd *d_cmd; void *cmd_data; rcu_id = srcu_read_lock(&einfo->use_ref); if (unlikely(!einfo->rx_fifo)) { if (!get_rx_fifo(einfo)) { pr_err("%s\trx fifo not found\n", glink_smem_native_driver.driver.name); srcu_read_unlock(&einfo->use_ref, rcu_id); return; } einfo->in_ssr = false; einfo->xprt_if.glink_core_if_ptr->link_up(&einfo->xprt_if); } if (einfo->in_ssr) { srcu_read_unlock(&einfo->use_ref, rcu_id); return; } if (!atomic_ctx) { if (einfo->tx_resume_needed && fifo_write_avail(einfo)) { einfo->tx_resume_needed = false; einfo->xprt_if.glink_core_if_ptr->tx_resume( &einfo->xprt_if); } spin_lock_irqsave(&einfo->write_lock, flags); if (waitqueue_active(&einfo->tx_blocked_queue)) { einfo->tx_blocked_signal_sent = false; trigger_wakeup = true; } spin_unlock_irqrestore(&einfo->write_lock, flags); if (trigger_wakeup) wake_up_all(&einfo->tx_blocked_queue); } /* * Access to the fifo needs to be synchronized, however only the calls * into the core from process_rx_data() are compatible with an atomic * processing context. For everything else, we need to do all the fifo * processing, then unlock the lock for the call into the core. Data * in the fifo is allowed to be processed immediately instead of being * ordered with the commands because the channel open process prevents * intents from being queued (which prevents data from being sent) until * all the channel open commands are processed by the core, thus * eliminating a race. */ spin_lock_irqsave(&einfo->rx_lock, flags); while (fifo_read_avail(einfo) || (!atomic_ctx && !list_empty(&einfo->deferred_cmds))) { if (einfo->in_ssr) break; if (!atomic_ctx && !list_empty(&einfo->deferred_cmds)) { d_cmd = list_first_entry(&einfo->deferred_cmds, struct deferred_cmd, list_node); list_del(&d_cmd->list_node); cmd.id = d_cmd->id; cmd.param1 = d_cmd->param1; cmd.param2 = d_cmd->param2; cmd_data = d_cmd->data; kfree(d_cmd); } else { fifo_read(einfo, &cmd, sizeof(cmd)); cmd_data = NULL; } switch (cmd.id) { case VERSION_CMD: if (atomic_ctx) { queue_cmd(einfo, &cmd, NULL); break; } spin_unlock_irqrestore(&einfo->rx_lock, flags); einfo->xprt_if.glink_core_if_ptr->rx_cmd_version( &einfo->xprt_if, cmd.param1, cmd.param2); spin_lock_irqsave(&einfo->rx_lock, flags); break; case VERSION_ACK_CMD: if (atomic_ctx) { queue_cmd(einfo, &cmd, NULL); break; } spin_unlock_irqrestore(&einfo->rx_lock, flags); einfo->xprt_if.glink_core_if_ptr->rx_cmd_version_ack( &einfo->xprt_if, cmd.param1, cmd.param2); spin_lock_irqsave(&einfo->rx_lock, flags); break; case OPEN_CMD: rcid = cmd.param1; name_len = cmd.param2; if (cmd_data) { name = cmd_data; } else { len = ALIGN(name_len, FIFO_ALIGNMENT); name = kmalloc(len, GFP_ATOMIC); if (!name) { pr_err("No memory available to rx ch open cmd name. Discarding cmd.\n"); while (len) { fifo_read(einfo, trash, FIFO_ALIGNMENT); len -= FIFO_ALIGNMENT; } break; } fifo_read(einfo, name, len); } if (atomic_ctx) { if (!queue_cmd(einfo, &cmd, name)) kfree(name); break; } spin_unlock_irqrestore(&einfo->rx_lock, flags); einfo->xprt_if.glink_core_if_ptr->rx_cmd_ch_remote_open( &einfo->xprt_if, rcid, name, SMEM_XPRT_ID); kfree(name); spin_lock_irqsave(&einfo->rx_lock, flags); break; case CLOSE_CMD: if (atomic_ctx) { queue_cmd(einfo, &cmd, NULL); break; } spin_unlock_irqrestore(&einfo->rx_lock, flags); einfo->xprt_if.glink_core_if_ptr-> rx_cmd_ch_remote_close( &einfo->xprt_if, cmd.param1); spin_lock_irqsave(&einfo->rx_lock, flags); break; case OPEN_ACK_CMD: if (atomic_ctx) { queue_cmd(einfo, &cmd, NULL); break; } spin_unlock_irqrestore(&einfo->rx_lock, flags); einfo->xprt_if.glink_core_if_ptr->rx_cmd_ch_open_ack( &einfo->xprt_if, cmd.param1, SMEM_XPRT_ID); spin_lock_irqsave(&einfo->rx_lock, flags); break; case RX_INTENT_CMD: /* * One intent listed with this command. This is the * expected case and can be optimized over the general * case of an array of intents. */ if (cmd.param2 == 1) { if (cmd_data) { intent.id = ((struct intent_desc *) cmd_data)->id; intent.size = ((struct intent_desc *) cmd_data)->size; kfree(cmd_data); } else { fifo_read(einfo, &intent, sizeof(intent)); } if (atomic_ctx) { cmd_data = kmalloc(sizeof(intent), GFP_ATOMIC); if (!cmd_data) { pr_err("%s: dropping cmd %d\n", __func__, cmd.id); break; } ((struct intent_desc *)cmd_data)->id = intent.id; ((struct intent_desc *)cmd_data)->size = intent.size; if (!queue_cmd(einfo, &cmd, cmd_data)) kfree(cmd_data); break; } spin_unlock_irqrestore(&einfo->rx_lock, flags); einfo->xprt_if.glink_core_if_ptr-> rx_cmd_remote_rx_intent_put( &einfo->xprt_if, cmd.param1, intent.id, intent.size); spin_lock_irqsave(&einfo->rx_lock, flags); break; } /* Array of intents to process */ if (cmd_data) { intents = cmd_data; } else { intents = kmalloc(sizeof(*intents) * cmd.param2, GFP_ATOMIC); if (!intents) { for (i = 0; i < cmd.param2; ++i) fifo_read(einfo, &intent, sizeof(intent)); break; } fifo_read(einfo, intents, sizeof(*intents) * cmd.param2); } if (atomic_ctx) { if (!queue_cmd(einfo, &cmd, intents)) kfree(intents); break; } spin_unlock_irqrestore(&einfo->rx_lock, flags); for (i = 0; i < cmd.param2; ++i) { einfo->xprt_if.glink_core_if_ptr-> rx_cmd_remote_rx_intent_put( &einfo->xprt_if, cmd.param1, intents[i].id, intents[i].size); } kfree(intents); spin_lock_irqsave(&einfo->rx_lock, flags); break; case RX_DONE_CMD: if (atomic_ctx) { queue_cmd(einfo, &cmd, NULL); break; } spin_unlock_irqrestore(&einfo->rx_lock, flags); einfo->xprt_if.glink_core_if_ptr->rx_cmd_tx_done( &einfo->xprt_if, cmd.param1, cmd.param2, false); spin_lock_irqsave(&einfo->rx_lock, flags); break; case RX_INTENT_REQ_CMD: if (atomic_ctx) { queue_cmd(einfo, &cmd, NULL); break; } spin_unlock_irqrestore(&einfo->rx_lock, flags); einfo->xprt_if.glink_core_if_ptr-> rx_cmd_remote_rx_intent_req( &einfo->xprt_if, cmd.param1, cmd.param2); spin_lock_irqsave(&einfo->rx_lock, flags); break; case RX_INTENT_REQ_ACK_CMD: if (atomic_ctx) { queue_cmd(einfo, &cmd, NULL); break; } spin_unlock_irqrestore(&einfo->rx_lock, flags); granted = false; if (cmd.param2 == 1) granted = true; einfo->xprt_if.glink_core_if_ptr-> rx_cmd_rx_intent_req_ack( &einfo->xprt_if, cmd.param1, granted); spin_lock_irqsave(&einfo->rx_lock, flags); break; case TX_DATA_CMD: case TX_DATA_CONT_CMD: case TRACER_PKT_CMD: case TRACER_PKT_CONT_CMD: process_rx_data(einfo, cmd.id, cmd.param1, cmd.param2); break; case CLOSE_ACK_CMD: if (atomic_ctx) { queue_cmd(einfo, &cmd, NULL); break; } spin_unlock_irqrestore(&einfo->rx_lock, flags); einfo->xprt_if.glink_core_if_ptr->rx_cmd_ch_close_ack( &einfo->xprt_if, cmd.param1); spin_lock_irqsave(&einfo->rx_lock, flags); break; case READ_NOTIF_CMD: send_irq(einfo); break; case SIGNALS_CMD: if (atomic_ctx) { queue_cmd(einfo, &cmd, NULL); break; } spin_unlock_irqrestore(&einfo->rx_lock, flags); einfo->xprt_if.glink_core_if_ptr->rx_cmd_remote_sigs( &einfo->xprt_if, cmd.param1, cmd.param2); spin_lock_irqsave(&einfo->rx_lock, flags); break; case RX_DONE_W_REUSE_CMD: if (atomic_ctx) { queue_cmd(einfo, &cmd, NULL); break; } spin_unlock_irqrestore(&einfo->rx_lock, flags); einfo->xprt_if.glink_core_if_ptr->rx_cmd_tx_done( &einfo->xprt_if, cmd.param1, cmd.param2, true); spin_lock_irqsave(&einfo->rx_lock, flags); break; default: pr_err("Unrecognized command: %d\n", cmd.id); break; } } spin_unlock_irqrestore(&einfo->rx_lock, flags); srcu_read_unlock(&einfo->use_ref, rcu_id); } /** * rx_worker() - worker function to process received commands * @work: kwork associated with the edge to process commands on. */ static void rx_worker(struct kthread_work *work) { struct edge_info *einfo; einfo = container_of(work, struct edge_info, kwork); __rx_worker(einfo, false); } irqreturn_t irq_handler(int irq, void *priv) { struct edge_info *einfo = (struct edge_info *)priv; if (einfo->rx_reset_reg) writel_relaxed(einfo->out_irq_mask, einfo->rx_reset_reg); queue_kthread_work(&einfo->kworker, &einfo->kwork); einfo->rx_irq_count++; return IRQ_HANDLED; } /** * tx_cmd_version() - convert a version cmd to wire format and transmit * @if_ptr: The transport to transmit on. * @version: The version number to encode. * @features: The features information to encode. */ static void tx_cmd_version(struct glink_transport_if *if_ptr, uint32_t version, uint32_t features) { struct command { uint16_t id; uint16_t version; uint32_t features; }; struct command cmd; struct edge_info *einfo; int rcu_id; einfo = container_of(if_ptr, struct edge_info, xprt_if); rcu_id = srcu_read_lock(&einfo->use_ref); if (einfo->in_ssr) { srcu_read_unlock(&einfo->use_ref, rcu_id); return; } cmd.id = VERSION_CMD; cmd.version = version; cmd.features = features; fifo_tx(einfo, &cmd, sizeof(cmd)); srcu_read_unlock(&einfo->use_ref, rcu_id); } /** * tx_cmd_version_ack() - convert a version ack cmd to wire format and transmit * @if_ptr: The transport to transmit on. * @version: The version number to encode. * @features: The features information to encode. */ static void tx_cmd_version_ack(struct glink_transport_if *if_ptr, uint32_t version, uint32_t features) { struct command { uint16_t id; uint16_t version; uint32_t features; }; struct command cmd; struct edge_info *einfo; int rcu_id; einfo = container_of(if_ptr, struct edge_info, xprt_if); rcu_id = srcu_read_lock(&einfo->use_ref); if (einfo->in_ssr) { srcu_read_unlock(&einfo->use_ref, rcu_id); return; } cmd.id = VERSION_ACK_CMD; cmd.version = version; cmd.features = features; fifo_tx(einfo, &cmd, sizeof(cmd)); srcu_read_unlock(&einfo->use_ref, rcu_id); } /** * set_version() - activate a negotiated version and feature set * @if_ptr: The transport to configure. * @version: The version to use. * @features: The features to use. * * Return: The supported capabilities of the transport. */ static uint32_t set_version(struct glink_transport_if *if_ptr, uint32_t version, uint32_t features) { struct edge_info *einfo; uint32_t ret; int rcu_id; einfo = container_of(if_ptr, struct edge_info, xprt_if); rcu_id = srcu_read_lock(&einfo->use_ref); if (einfo->in_ssr) { srcu_read_unlock(&einfo->use_ref, rcu_id); return 0; } ret = einfo->intentless ? GCAP_INTENTLESS | GCAP_SIGNALS : GCAP_SIGNALS; if (features & TRACER_PKT_FEATURE) ret |= GCAP_TRACER_PKT; srcu_read_unlock(&einfo->use_ref, rcu_id); return ret; } /** * tx_cmd_ch_open() - convert a channel open cmd to wire format and transmit * @if_ptr: The transport to transmit on. * @lcid: The local channel id to encode. * @name: The channel name to encode. * @req_xprt: The transport the core would like to migrate this channel to. * * Return: 0 on success or standard Linux error code. */ static int tx_cmd_ch_open(struct glink_transport_if *if_ptr, uint32_t lcid, const char *name, uint16_t req_xprt) { struct command { uint16_t id; uint16_t lcid; uint32_t length; }; struct command cmd; struct edge_info *einfo; uint32_t buf_size; void *buf; int rcu_id; einfo = container_of(if_ptr, struct edge_info, xprt_if); rcu_id = srcu_read_lock(&einfo->use_ref); if (einfo->in_ssr) { srcu_read_unlock(&einfo->use_ref, rcu_id); return -EFAULT; } cmd.id = OPEN_CMD; cmd.lcid = lcid; cmd.length = strlen(name) + 1; buf_size = ALIGN(sizeof(cmd) + cmd.length, FIFO_ALIGNMENT); buf = kzalloc(buf_size, GFP_KERNEL); if (!buf) { pr_err("%s: malloc fail for %d size buf\n", __func__, buf_size); srcu_read_unlock(&einfo->use_ref, rcu_id); return -ENOMEM; } memcpy(buf, &cmd, sizeof(cmd)); memcpy(buf + sizeof(cmd), name, cmd.length); fifo_tx(einfo, buf, buf_size); kfree(buf); srcu_read_unlock(&einfo->use_ref, rcu_id); return 0; } /** * tx_cmd_ch_close() - convert a channel close cmd to wire format and transmit * @if_ptr: The transport to transmit on. * @lcid: The local channel id to encode. * * Return: 0 on success or standard Linux error code. */ static int tx_cmd_ch_close(struct glink_transport_if *if_ptr, uint32_t lcid) { struct command { uint16_t id; uint16_t lcid; uint32_t reserved; }; struct command cmd; struct edge_info *einfo; int rcu_id; einfo = container_of(if_ptr, struct edge_info, xprt_if); rcu_id = srcu_read_lock(&einfo->use_ref); if (einfo->in_ssr) { srcu_read_unlock(&einfo->use_ref, rcu_id); return -EFAULT; } cmd.id = CLOSE_CMD; cmd.lcid = lcid; cmd.reserved = 0; fifo_tx(einfo, &cmd, sizeof(cmd)); srcu_read_unlock(&einfo->use_ref, rcu_id); return 0; } /** * tx_cmd_ch_remote_open_ack() - convert a channel open ack cmd to wire format * and transmit * @if_ptr: The transport to transmit on. * @rcid: The remote channel id to encode. * @xprt_resp: The response to a transport migration request. */ static void tx_cmd_ch_remote_open_ack(struct glink_transport_if *if_ptr, uint32_t rcid, uint16_t xprt_resp) { struct command { uint16_t id; uint16_t rcid; uint32_t reserved; }; struct command cmd; struct edge_info *einfo; int rcu_id; einfo = container_of(if_ptr, struct edge_info, xprt_if); rcu_id = srcu_read_lock(&einfo->use_ref); if (einfo->in_ssr) { srcu_read_unlock(&einfo->use_ref, rcu_id); return; } cmd.id = OPEN_ACK_CMD; cmd.rcid = rcid; cmd.reserved = 0; fifo_tx(einfo, &cmd, sizeof(cmd)); srcu_read_unlock(&einfo->use_ref, rcu_id); } /** * tx_cmd_ch_remote_close_ack() - convert a channel close ack cmd to wire format * and transmit * @if_ptr: The transport to transmit on. * @rcid: The remote channel id to encode. */ static void tx_cmd_ch_remote_close_ack(struct glink_transport_if *if_ptr, uint32_t rcid) { struct command { uint16_t id; uint16_t rcid; uint32_t reserved; }; struct command cmd; struct edge_info *einfo; int rcu_id; einfo = container_of(if_ptr, struct edge_info, xprt_if); rcu_id = srcu_read_lock(&einfo->use_ref); if (einfo->in_ssr) { srcu_read_unlock(&einfo->use_ref, rcu_id); return; } cmd.id = CLOSE_ACK_CMD; cmd.rcid = rcid; cmd.reserved = 0; fifo_tx(einfo, &cmd, sizeof(cmd)); srcu_read_unlock(&einfo->use_ref, rcu_id); } /** * int wait_link_down() - Check status of read/write indices * @if_ptr: The transport to check * * Return: 1 if indices are all zero, 0 otherwise */ int wait_link_down(struct glink_transport_if *if_ptr) { struct edge_info *einfo; einfo = container_of(if_ptr, struct edge_info, xprt_if); if (einfo->tx_ch_desc->write_index == 0 && einfo->tx_ch_desc->read_index == 0 && einfo->rx_ch_desc->write_index == 0 && einfo->rx_ch_desc->read_index == 0) return 1; else return 0; } /** * allocate_rx_intent() - allocate/reserve space for RX Intent * @if_ptr: The transport the intent is associated with. * @size: size of intent. * @intent: Pointer to the intent structure. * * Assign "data" with the buffer created, since the transport creates * a linear buffer and "iovec" with the "intent" itself, so that * the data can be passed to a client that receives only vector buffer. * Note that returning NULL for the pointer is valid (it means that space has * been reserved, but the actual pointer will be provided later). * * Return: 0 on success or standard Linux error code. */ static int allocate_rx_intent(struct glink_transport_if *if_ptr, size_t size, struct glink_core_rx_intent *intent) { void *t; t = kmalloc(size, GFP_KERNEL); if (!t) return -ENOMEM; intent->data = t; intent->iovec = (void *)intent; intent->vprovider = rx_linear_vbuf_provider; intent->pprovider = NULL; return 0; } /** * deallocate_rx_intent() - Deallocate space created for RX Intent * @if_ptr: The transport the intent is associated with. * @intent: Pointer to the intent structure. * * Return: 0 on success or standard Linux error code. */ static int deallocate_rx_intent(struct glink_transport_if *if_ptr, struct glink_core_rx_intent *intent) { if (!intent || !intent->data) return -EINVAL; kfree(intent->data); intent->data = NULL; intent->iovec = NULL; intent->vprovider = NULL; return 0; } /** * tx_cmd_local_rx_intent() - convert an rx intent cmd to wire format and * transmit * @if_ptr: The transport to transmit on. * @lcid: The local channel id to encode. * @size: The intent size to encode. * @liid: The local intent id to encode. * * Return: 0 on success or standard Linux error code. */ static int tx_cmd_local_rx_intent(struct glink_transport_if *if_ptr, uint32_t lcid, size_t size, uint32_t liid) { struct command { uint16_t id; uint16_t lcid; uint32_t count; uint32_t size; uint32_t liid; }; struct command cmd; struct edge_info *einfo; int rcu_id; if (size > UINT_MAX) { pr_err("%s: size %zu is too large to encode\n", __func__, size); return -EMSGSIZE; } einfo = container_of(if_ptr, struct edge_info, xprt_if); if (einfo->intentless) return -EOPNOTSUPP; rcu_id = srcu_read_lock(&einfo->use_ref); if (einfo->in_ssr) { srcu_read_unlock(&einfo->use_ref, rcu_id); return -EFAULT; } cmd.id = RX_INTENT_CMD; cmd.lcid = lcid; cmd.count = 1; cmd.size = size; cmd.liid = liid; fifo_tx(einfo, &cmd, sizeof(cmd)); srcu_read_unlock(&einfo->use_ref, rcu_id); return 0; } /** * tx_cmd_local_rx_done() - convert an rx done cmd to wire format and transmit * @if_ptr: The transport to transmit on. * @lcid: The local channel id to encode. * @liid: The local intent id to encode. * @reuse: Reuse the consumed intent. */ static void tx_cmd_local_rx_done(struct glink_transport_if *if_ptr, uint32_t lcid, uint32_t liid, bool reuse) { struct command { uint16_t id; uint16_t lcid; uint32_t liid; }; struct command cmd; struct edge_info *einfo; int rcu_id; einfo = container_of(if_ptr, struct edge_info, xprt_if); if (einfo->intentless) return; rcu_id = srcu_read_lock(&einfo->use_ref); if (einfo->in_ssr) { srcu_read_unlock(&einfo->use_ref, rcu_id); return; } cmd.id = reuse ? RX_DONE_W_REUSE_CMD : RX_DONE_CMD; cmd.lcid = lcid; cmd.liid = liid; fifo_tx(einfo, &cmd, sizeof(cmd)); srcu_read_unlock(&einfo->use_ref, rcu_id); } /** * tx_cmd_rx_intent_req() - convert an rx intent request cmd to wire format and * transmit * @if_ptr: The transport to transmit on. * @lcid: The local channel id to encode. * @size: The requested intent size to encode. * * Return: 0 on success or standard Linux error code. */ static int tx_cmd_rx_intent_req(struct glink_transport_if *if_ptr, uint32_t lcid, size_t size) { struct command { uint16_t id; uint16_t lcid; uint32_t size; }; struct command cmd; struct edge_info *einfo; int rcu_id; if (size > UINT_MAX) { pr_err("%s: size %zu is too large to encode\n", __func__, size); return -EMSGSIZE; } einfo = container_of(if_ptr, struct edge_info, xprt_if); if (einfo->intentless) return -EOPNOTSUPP; rcu_id = srcu_read_lock(&einfo->use_ref); if (einfo->in_ssr) { srcu_read_unlock(&einfo->use_ref, rcu_id); return -EFAULT; } cmd.id = RX_INTENT_REQ_CMD, cmd.lcid = lcid; cmd.size = size; fifo_tx(einfo, &cmd, sizeof(cmd)); srcu_read_unlock(&einfo->use_ref, rcu_id); return 0; } /** * tx_cmd_rx_intent_req_ack() - convert an rx intent request ack cmd to wire * format and transmit * @if_ptr: The transport to transmit on. * @lcid: The local channel id to encode. * @granted: The request response to encode. * * Return: 0 on success or standard Linux error code. */ static int tx_cmd_remote_rx_intent_req_ack(struct glink_transport_if *if_ptr, uint32_t lcid, bool granted) { struct command { uint16_t id; uint16_t lcid; uint32_t response; }; struct command cmd; struct edge_info *einfo; int rcu_id; einfo = container_of(if_ptr, struct edge_info, xprt_if); if (einfo->intentless) return -EOPNOTSUPP; rcu_id = srcu_read_lock(&einfo->use_ref); if (einfo->in_ssr) { srcu_read_unlock(&einfo->use_ref, rcu_id); return -EFAULT; } cmd.id = RX_INTENT_REQ_ACK_CMD, cmd.lcid = lcid; if (granted) cmd.response = 1; else cmd.response = 0; fifo_tx(einfo, &cmd, sizeof(cmd)); srcu_read_unlock(&einfo->use_ref, rcu_id); return 0; } /** * tx_cmd_set_sigs() - convert a signals ack cmd to wire format and transmit * @if_ptr: The transport to transmit on. * @lcid: The local channel id to encode. * @sigs: The signals to encode. * * Return: 0 on success or standard Linux error code. */ static int tx_cmd_set_sigs(struct glink_transport_if *if_ptr, uint32_t lcid, uint32_t sigs) { struct command { uint16_t id; uint16_t lcid; uint32_t sigs; }; struct command cmd; struct edge_info *einfo; int rcu_id; einfo = container_of(if_ptr, struct edge_info, xprt_if); rcu_id = srcu_read_lock(&einfo->use_ref); if (einfo->in_ssr) { srcu_read_unlock(&einfo->use_ref, rcu_id); return -EFAULT; } cmd.id = SIGNALS_CMD, cmd.lcid = lcid; cmd.sigs = sigs; fifo_tx(einfo, &cmd, sizeof(cmd)); srcu_read_unlock(&einfo->use_ref, rcu_id); return 0; } /** * poll() - poll for data on a channel * @if_ptr: The transport the channel exists on. * @lcid: The local channel id. * * Return: 0 if no data available, 1 if data available. */ static int poll(struct glink_transport_if *if_ptr, uint32_t lcid) { struct edge_info *einfo; int rcu_id; einfo = container_of(if_ptr, struct edge_info, xprt_if); rcu_id = srcu_read_lock(&einfo->use_ref); if (einfo->in_ssr) { srcu_read_unlock(&einfo->use_ref, rcu_id); return -EFAULT; } if (fifo_read_avail(einfo)) { __rx_worker(einfo, true); srcu_read_unlock(&einfo->use_ref, rcu_id); return 1; } srcu_read_unlock(&einfo->use_ref, rcu_id); return 0; } /** * mask_rx_irq() - mask the receive irq for a channel * @if_ptr: The transport the channel exists on. * @lcid: The local channel id for the channel. * @mask: True to mask the irq, false to unmask. * @pstruct: Platform defined structure for handling the masking. * * Return: 0 on success or standard Linux error code. */ static int mask_rx_irq(struct glink_transport_if *if_ptr, uint32_t lcid, bool mask, void *pstruct) { struct edge_info *einfo; struct irq_chip *irq_chip; struct irq_data *irq_data; int rcu_id; einfo = container_of(if_ptr, struct edge_info, xprt_if); rcu_id = srcu_read_lock(&einfo->use_ref); if (einfo->in_ssr) { srcu_read_unlock(&einfo->use_ref, rcu_id); return -EFAULT; } irq_chip = irq_get_chip(einfo->irq_line); if (!irq_chip) { srcu_read_unlock(&einfo->use_ref, rcu_id); return -ENODEV; } irq_data = irq_get_irq_data(einfo->irq_line); if (!irq_data) { srcu_read_unlock(&einfo->use_ref, rcu_id); return -ENODEV; } if (mask) { irq_chip->irq_mask(irq_data); einfo->irq_disabled = true; if (pstruct) irq_set_affinity(einfo->irq_line, pstruct); } else { irq_chip->irq_unmask(irq_data); einfo->irq_disabled = false; } srcu_read_unlock(&einfo->use_ref, rcu_id); return 0; } /** * tx_data() - convert a data/tracer_pkt to wire format and transmit * @if_ptr: The transport to transmit on. * @cmd_id: The command ID to transmit. * @lcid: The local channel id to encode. * @pctx: The data to encode. * * Return: Number of bytes written or standard Linux error code. */ static int tx_data(struct glink_transport_if *if_ptr, uint16_t cmd_id, uint32_t lcid, struct glink_core_tx_pkt *pctx) { struct command { uint16_t id; uint16_t lcid; uint32_t riid; uint32_t size; uint32_t size_left; }; struct command cmd; struct edge_info *einfo; uint32_t size; uint32_t zeros_size; const void *data_start; char zeros[FIFO_ALIGNMENT] = { 0 }; unsigned long flags; size_t tx_size = 0; int rcu_id; if (pctx->size < pctx->size_remaining) { GLINK_ERR("%s: size remaining exceeds size. Resetting.\n", __func__); pctx->size_remaining = pctx->size; } if (!pctx->size_remaining) return 0; einfo = container_of(if_ptr, struct edge_info, xprt_if); rcu_id = srcu_read_lock(&einfo->use_ref); if (einfo->in_ssr) { srcu_read_unlock(&einfo->use_ref, rcu_id); return -EFAULT; } if (einfo->intentless && (pctx->size_remaining != pctx->size || cmd_id == TRACER_PKT_CMD)) { srcu_read_unlock(&einfo->use_ref, rcu_id); return -EINVAL; } if (cmd_id == TX_DATA_CMD) { if (pctx->size_remaining == pctx->size) cmd.id = TX_DATA_CMD; else cmd.id = TX_DATA_CONT_CMD; } else { if (pctx->size_remaining == pctx->size) cmd.id = TRACER_PKT_CMD; else cmd.id = TRACER_PKT_CONT_CMD; } cmd.lcid = lcid; cmd.riid = pctx->riid; data_start = get_tx_vaddr(pctx, pctx->size - pctx->size_remaining, &tx_size); if (!data_start) { GLINK_ERR("%s: invalid data_start\n", __func__); srcu_read_unlock(&einfo->use_ref, rcu_id); return -EINVAL; } spin_lock_irqsave(&einfo->write_lock, flags); size = fifo_write_avail(einfo); /* Intentless clients expect a complete commit or instant failure */ if (einfo->intentless && size < sizeof(cmd) + pctx->size) { spin_unlock_irqrestore(&einfo->write_lock, flags); srcu_read_unlock(&einfo->use_ref, rcu_id); return -ENOSPC; } /* Need enough space to write the command and some data */ if (size <= sizeof(cmd)) { einfo->tx_resume_needed = true; spin_unlock_irqrestore(&einfo->write_lock, flags); srcu_read_unlock(&einfo->use_ref, rcu_id); return -EAGAIN; } size -= sizeof(cmd); if (size > tx_size) size = tx_size; cmd.size = size; pctx->size_remaining -= size; cmd.size_left = pctx->size_remaining; zeros_size = ALIGN(size, FIFO_ALIGNMENT) - cmd.size; if (cmd.id == TRACER_PKT_CMD) tracer_pkt_log_event((void *)(pctx->data), GLINK_XPRT_TX); fifo_write_complex(einfo, &cmd, sizeof(cmd), data_start, size, zeros, zeros_size); GLINK_DBG("%s %s: lcid[%u] riid[%u] cmd[%d], size[%d], size_left[%d]\n", "", __func__, cmd.lcid, cmd.riid, cmd.id, cmd.size, cmd.size_left); spin_unlock_irqrestore(&einfo->write_lock, flags); /* Fake tx_done for intentless since its not supported over the wire */ if (einfo->intentless) { spin_lock_irqsave(&einfo->rx_lock, flags); cmd.id = RX_DONE_CMD; cmd.lcid = pctx->rcid; queue_cmd(einfo, &cmd, NULL); spin_unlock_irqrestore(&einfo->rx_lock, flags); } srcu_read_unlock(&einfo->use_ref, rcu_id); return cmd.size; } /** * tx() - convert a data transmit cmd to wire format and transmit * @if_ptr: The transport to transmit on. * @lcid: The local channel id to encode. * @pctx: The data to encode. * * Return: Number of bytes written or standard Linux error code. */ static int tx(struct glink_transport_if *if_ptr, uint32_t lcid, struct glink_core_tx_pkt *pctx) { return tx_data(if_ptr, TX_DATA_CMD, lcid, pctx); } /** * tx_cmd_tracer_pkt() - convert a tracer packet cmd to wire format and transmit * @if_ptr: The transport to transmit on. * @lcid: The local channel id to encode. * @pctx: The data to encode. * * Return: Number of bytes written or standard Linux error code. */ static int tx_cmd_tracer_pkt(struct glink_transport_if *if_ptr, uint32_t lcid, struct glink_core_tx_pkt *pctx) { return tx_data(if_ptr, TRACER_PKT_CMD, lcid, pctx); } /** * get_power_vote_ramp_time() - Get the ramp time required for the power * votes to be applied * @if_ptr: The transport interface on which power voting is requested. * @state: The power state for which ramp time is required. * * Return: The ramp time specific to the power state, standard error otherwise. */ static unsigned long get_power_vote_ramp_time( struct glink_transport_if *if_ptr, uint32_t state) { struct edge_info *einfo; einfo = container_of(if_ptr, struct edge_info, xprt_if); if (state >= einfo->num_pw_states || !(einfo->ramp_time_us)) return (unsigned long)ERR_PTR(-EINVAL); return einfo->ramp_time_us[state]; } /** * power_vote() - Update the power votes to meet qos requirement * @if_ptr: The transport interface on which power voting is requested. * @state: The power state for which the voting should be done. * * Return: 0 on Success, standard error otherwise. */ static int power_vote(struct glink_transport_if *if_ptr, uint32_t state) { return 0; } /** * power_unvote() - Remove the all the power votes * @if_ptr: The transport interface on which power voting is requested. * * Return: 0 on Success, standard error otherwise. */ static int power_unvote(struct glink_transport_if *if_ptr) { return 0; } /** * negotiate_features_v1() - determine what features of a version can be used * @if_ptr: The transport for which features are negotiated for. * @version: The version negotiated. * @features: The set of requested features. * * Return: What set of the requested features can be supported. */ static uint32_t negotiate_features_v1(struct glink_transport_if *if_ptr, const struct glink_core_version *version, uint32_t features) { return features & version->features; } /** * init_xprt_if() - initialize the xprt_if for an edge * @einfo: The edge to initialize. */ static void init_xprt_if(struct edge_info *einfo) { einfo->xprt_if.tx_cmd_version = tx_cmd_version; einfo->xprt_if.tx_cmd_version_ack = tx_cmd_version_ack; einfo->xprt_if.set_version = set_version; einfo->xprt_if.tx_cmd_ch_open = tx_cmd_ch_open; einfo->xprt_if.tx_cmd_ch_close = tx_cmd_ch_close; einfo->xprt_if.tx_cmd_ch_remote_open_ack = tx_cmd_ch_remote_open_ack; einfo->xprt_if.tx_cmd_ch_remote_close_ack = tx_cmd_ch_remote_close_ack; einfo->xprt_if.ssr = ssr; einfo->xprt_if.allocate_rx_intent = allocate_rx_intent; einfo->xprt_if.deallocate_rx_intent = deallocate_rx_intent; einfo->xprt_if.tx_cmd_local_rx_intent = tx_cmd_local_rx_intent; einfo->xprt_if.tx_cmd_local_rx_done = tx_cmd_local_rx_done; einfo->xprt_if.tx = tx; einfo->xprt_if.tx_cmd_rx_intent_req = tx_cmd_rx_intent_req; einfo->xprt_if.tx_cmd_remote_rx_intent_req_ack = tx_cmd_remote_rx_intent_req_ack; einfo->xprt_if.tx_cmd_set_sigs = tx_cmd_set_sigs; einfo->xprt_if.poll = poll; einfo->xprt_if.mask_rx_irq = mask_rx_irq; einfo->xprt_if.wait_link_down = wait_link_down; einfo->xprt_if.tx_cmd_tracer_pkt = tx_cmd_tracer_pkt; einfo->xprt_if.get_power_vote_ramp_time = get_power_vote_ramp_time; einfo->xprt_if.power_vote = power_vote; einfo->xprt_if.power_unvote = power_unvote; } /** * init_xprt_cfg() - initialize the xprt_cfg for an edge * @einfo: The edge to initialize. * @name: The name of the remote side this edge communicates to. */ static void init_xprt_cfg(struct edge_info *einfo, const char *name) { einfo->xprt_cfg.name = XPRT_NAME; einfo->xprt_cfg.edge = name; einfo->xprt_cfg.versions = versions; einfo->xprt_cfg.versions_entries = ARRAY_SIZE(versions); einfo->xprt_cfg.max_cid = SZ_64K; einfo->xprt_cfg.max_iid = SZ_2G; } /** * parse_qos_dt_params() - Parse the power states from DT * @dev: Reference to the platform device for a specific edge. * @einfo: Edge information for the edge probe function is called. * * Return: 0 on success, standard error code otherwise. */ static int parse_qos_dt_params(struct device_node *node, struct edge_info *einfo) { int rc; int i; char *key; uint32_t *arr32; uint32_t num_states; key = "qcom,ramp-time"; if (!of_find_property(node, key, &num_states)) return -ENODEV; num_states /= sizeof(uint32_t); einfo->num_pw_states = num_states; arr32 = kmalloc_array(num_states, sizeof(uint32_t), GFP_KERNEL); if (!arr32) return -ENOMEM; einfo->ramp_time_us = kmalloc_array(num_states, sizeof(unsigned long), GFP_KERNEL); if (!einfo->ramp_time_us) { rc = -ENOMEM; goto mem_alloc_fail; } rc = of_property_read_u32_array(node, key, arr32, num_states); if (rc) { rc = -ENODEV; goto invalid_key; } for (i = 0; i < num_states; i++) einfo->ramp_time_us[i] = arr32[i]; rc = 0; return rc; invalid_key: kfree(einfo->ramp_time_us); mem_alloc_fail: kfree(arr32); return rc; } void get_entry_by_name(struct device_node *node, char *id, uint32_t *value) { int index = 0; if (id) index = of_property_match_string(node, "smem-entry-names", id); of_property_read_u32_index(node, "smem-entry", index, value); } static int glink_smem_native_probe(struct platform_device *pdev) { struct device_node *node; struct device_node *phandle_node; struct edge_info *einfo; int rc; char *key; const char *subsys_name; uint32_t irq_line; uint32_t irq_mask; uint32_t ch_desc_size = SMEM_CH_DESC_SIZE; uint32_t entry = 0; uint32_t remote_proc_id; struct resource *r; node = pdev->dev.of_node; key = "qcom,subsys-id"; rc = of_property_read_u32(node, key, &remote_proc_id); if (rc) { pr_err("%s: missing key %s\n", __func__, key); return -ENODEV; } key = "ch"; get_entry_by_name(node, key, &entry); rc = qcom_smem_alloc(remote_proc_id, entry, ch_desc_size); if ((rc != -EEXIST) && (rc != 0)) return -EPROBE_DEFER; einfo = kzalloc(sizeof(*einfo), GFP_KERNEL); if (!einfo) { pr_err("%s: edge_info allocation failed\n", __func__); rc = -ENOMEM; goto edge_info_alloc_fail; } einfo->remote_proc_id = remote_proc_id; key = "label"; subsys_name = of_get_property(node, key, NULL); if (!subsys_name) { pr_err("%s: missing key %s\n", __func__, key); rc = -ENODEV; goto missing_key; } key = "interrupts"; irq_line = irq_of_parse_and_map(node, 0); if (!irq_line) { pr_err("%s: missing key %s\n", __func__, key); rc = -ENODEV; goto missing_key; } key = "qcom,irq-mask"; rc = of_property_read_u32(node, key, &irq_mask); if (rc) { pr_err("%s: missing key %s\n", __func__, key); rc = -ENODEV; goto missing_key; } key = "irq-reg-base"; r = platform_get_resource_byname(pdev, IORESOURCE_MEM, key); if (!r) { pr_err("%s: missing key %s\n", __func__, key); rc = -ENODEV; goto missing_key; } init_xprt_cfg(einfo, subsys_name); init_xprt_if(einfo); spin_lock_init(&einfo->write_lock); init_waitqueue_head(&einfo->tx_blocked_queue); init_kthread_work(&einfo->kwork, rx_worker); init_kthread_worker(&einfo->kworker); einfo->read_from_fifo = read_from_fifo; einfo->write_to_fifo = write_to_fifo; init_srcu_struct(&einfo->use_ref); spin_lock_init(&einfo->rx_lock); INIT_LIST_HEAD(&einfo->deferred_cmds); mutex_lock(&probe_lock); if (edge_infos[einfo->remote_proc_id]) { pr_err("%s: duplicate subsys %s is not valid\n", __func__, subsys_name); rc = -ENODEV; mutex_unlock(&probe_lock); goto invalid_key; } edge_infos[einfo->remote_proc_id] = einfo; mutex_unlock(&probe_lock); einfo->out_irq_mask = irq_mask; einfo->out_irq_reg = ioremap_nocache(r->start, resource_size(r)); if (!einfo->out_irq_reg) { pr_err("%s: unable to map irq reg\n", __func__); rc = -ENOMEM; goto ioremap_fail; } einfo->task = kthread_run(kthread_worker_fn, &einfo->kworker, "smem_native_%s", subsys_name); if (IS_ERR(einfo->task)) { rc = PTR_ERR(einfo->task); pr_err("%s: kthread_run failed %d\n", __func__, rc); goto kthread_fail; } einfo->tx_ch_desc = qcom_smem_get(einfo->remote_proc_id, entry, &ch_desc_size); if (PTR_ERR(einfo->tx_ch_desc) == -EPROBE_DEFER) { goto smem_alloc_fail; } if (!einfo->tx_ch_desc) { pr_err("%s: smem retrieval of ch descriptor failed\n", __func__); goto smem_alloc_fail; } einfo->rx_ch_desc = einfo->tx_ch_desc + 1; einfo->tx_fifo_size = SZ_16K; key = "tx_fifo"; get_entry_by_name(node, key, &einfo->tx_fifo_number); rc = qcom_smem_alloc(einfo->remote_proc_id, einfo->tx_fifo_number, einfo->tx_fifo_size); if (rc) { pr_err("%s: smem alloc of ch descriptor failed\n", __func__); goto smem_alloc_fail; } einfo->tx_fifo = qcom_smem_get(einfo->remote_proc_id, einfo->tx_fifo_number, &einfo->tx_fifo_size); if (!einfo->tx_fifo) { pr_err("%s: smem retrieval of tx fifo failed\n", __func__); goto smem_alloc_fail; } key = "rx_fifo"; get_entry_by_name(node, key, &einfo->rx_fifo_number); key = "qcom,qos-config"; phandle_node = of_parse_phandle(node, key, 0); if (phandle_node && !(of_get_glink_core_qos_cfg(phandle_node, &einfo->xprt_cfg))) parse_qos_dt_params(node, einfo); rc = glink_core_register_transport(&einfo->xprt_if, &einfo->xprt_cfg); if (rc == -EPROBE_DEFER) goto reg_xprt_fail; if (rc) { pr_err("%s: glink core register transport failed: %d\n", __func__, rc); goto reg_xprt_fail; } einfo->irq_line = irq_line; rc = request_irq(irq_line, irq_handler, IRQF_TRIGGER_RISING | IRQF_NO_SUSPEND | IRQF_SHARED, node->name, einfo); if (rc < 0) { pr_err("%s: request_irq on %d failed: %d\n", __func__, irq_line, rc); goto request_irq_fail; } rc = enable_irq_wake(irq_line); if (rc < 0) pr_err("%s: enable_irq_wake() failed on %d\n", __func__, irq_line); register_debugfs_info(einfo); /* fake an interrupt on this edge to see if the remote side is up */ irq_handler(0, einfo); return 0; request_irq_fail: glink_core_unregister_transport(&einfo->xprt_if); reg_xprt_fail: smem_alloc_fail: flush_kthread_worker(&einfo->kworker); kthread_stop(einfo->task); einfo->task = NULL; kthread_fail: iounmap(einfo->out_irq_reg); ioremap_fail: mutex_lock(&probe_lock); edge_infos[einfo->remote_proc_id] = NULL; mutex_unlock(&probe_lock); invalid_key: missing_key: kfree(einfo); edge_info_alloc_fail: return rc; } static int glink_rpm_native_probe(struct platform_device *pdev) { struct device_node *node; struct edge_info *einfo; int rc; char *key; const char *subsys_name; uint32_t irq_line; uint32_t irq_mask; struct resource *irq_r; struct resource *msgram_r; void __iomem *msgram; char toc[RPM_TOC_SIZE]; uint32_t *tocp; uint32_t num_toc_entries; node = pdev->dev.of_node; einfo = kzalloc(sizeof(*einfo), GFP_KERNEL); if (!einfo) { pr_err("%s: edge_info allocation failed\n", __func__); rc = -ENOMEM; goto edge_info_alloc_fail; } subsys_name = "rpm"; key = "interrupts"; irq_line = irq_of_parse_and_map(node, 0); if (!irq_line) { pr_err("%s: missing key %s\n", __func__, key); rc = -ENODEV; goto missing_key; } key = "qcom,irq-mask"; rc = of_property_read_u32(node, key, &irq_mask); if (rc) { pr_err("%s: missing key %s\n", __func__, key); rc = -ENODEV; goto missing_key; } key = "irq-reg-base"; irq_r = platform_get_resource_byname(pdev, IORESOURCE_MEM, key); if (!irq_r) { pr_err("%s: missing key %s\n", __func__, key); rc = -ENODEV; goto missing_key; } key = "msgram"; msgram_r = platform_get_resource_byname(pdev, IORESOURCE_MEM, key); if (!msgram_r) { pr_err("%s: missing key %s\n", __func__, key); rc = -ENODEV; goto missing_key; } key = "qcom,subsys-id"; rc = of_property_read_u32(node, key, &einfo->remote_proc_id); if (rc) { pr_err("%s: missing key %s\n", __func__, key); rc = -ENODEV; goto missing_key; } init_xprt_cfg(einfo, subsys_name); init_xprt_if(einfo); spin_lock_init(&einfo->write_lock); init_waitqueue_head(&einfo->tx_blocked_queue); init_kthread_work(&einfo->kwork, rx_worker); init_kthread_worker(&einfo->kworker); einfo->intentless = true; einfo->read_from_fifo = memcpy32_fromio; einfo->write_to_fifo = memcpy32_toio; init_srcu_struct(&einfo->use_ref); spin_lock_init(&einfo->rx_lock); INIT_LIST_HEAD(&einfo->deferred_cmds); mutex_lock(&probe_lock); if (edge_infos[einfo->remote_proc_id]) { pr_err("%s: duplicate subsys %s is not valid\n", __func__, subsys_name); rc = -ENODEV; mutex_unlock(&probe_lock); goto invalid_key; } edge_infos[einfo->remote_proc_id] = einfo; mutex_unlock(&probe_lock); einfo->out_irq_mask = irq_mask; einfo->out_irq_reg = ioremap_nocache(irq_r->start, resource_size(irq_r)); if (!einfo->out_irq_reg) { pr_err("%s: unable to map irq reg\n", __func__); rc = -ENOMEM; goto irq_ioremap_fail; } msgram = ioremap_nocache(msgram_r->start, resource_size(msgram_r)); if (!msgram) { pr_err("%s: unable to map msgram\n", __func__); rc = -ENOMEM; goto msgram_ioremap_fail; } einfo->task = kthread_run(kthread_worker_fn, &einfo->kworker, "smem_native_%s", subsys_name); if (IS_ERR(einfo->task)) { rc = PTR_ERR(einfo->task); pr_err("%s: kthread_run failed %d\n", __func__, rc); goto kthread_fail; } memcpy32_fromio(toc, msgram + resource_size(msgram_r) - RPM_TOC_SIZE, RPM_TOC_SIZE); tocp = (uint32_t *)toc; if (*tocp != RPM_TOC_ID) { rc = -ENODEV; pr_err("%s: TOC id %d is not valid\n", __func__, *tocp); goto toc_init_fail; } ++tocp; num_toc_entries = *tocp; if (num_toc_entries > RPM_MAX_TOC_ENTRIES) { rc = -ENODEV; pr_err("%s: %d is too many toc entries\n", __func__, num_toc_entries); goto toc_init_fail; } ++tocp; for (rc = 0; rc < num_toc_entries; ++rc) { if (*tocp != RPM_TX_FIFO_ID) { tocp += 3; continue; } ++tocp; einfo->tx_ch_desc = msgram + *tocp; einfo->tx_fifo = einfo->tx_ch_desc + 1; if ((uintptr_t)einfo->tx_fifo > (uintptr_t)(msgram + resource_size(msgram_r))) { pr_err("%s: invalid tx fifo address\n", __func__); einfo->tx_fifo = NULL; break; } ++tocp; einfo->tx_fifo_size = *tocp; if (einfo->tx_fifo_size > resource_size(msgram_r) || (uintptr_t)(einfo->tx_fifo + einfo->tx_fifo_size) > (uintptr_t)(msgram + resource_size(msgram_r))) { pr_err("%s: invalid tx fifo size\n", __func__); einfo->tx_fifo = NULL; break; } break; } if (!einfo->tx_fifo) { rc = -ENODEV; pr_err("%s: tx fifo not found\n", __func__); goto toc_init_fail; } tocp = (uint32_t *)toc; tocp += 2; for (rc = 0; rc < num_toc_entries; ++rc) { if (*tocp != RPM_RX_FIFO_ID) { tocp += 3; continue; } ++tocp; einfo->rx_ch_desc = msgram + *tocp; einfo->rx_fifo = einfo->rx_ch_desc + 1; if ((uintptr_t)einfo->rx_fifo > (uintptr_t)(msgram + resource_size(msgram_r))) { pr_err("%s: invalid rx fifo address\n", __func__); einfo->rx_fifo = NULL; break; } ++tocp; einfo->rx_fifo_size = *tocp; if (einfo->rx_fifo_size > resource_size(msgram_r) || (uintptr_t)(einfo->rx_fifo + einfo->rx_fifo_size) > (uintptr_t)(msgram + resource_size(msgram_r))) { pr_err("%s: invalid rx fifo size\n", __func__); einfo->rx_fifo = NULL; break; } break; } if (!einfo->rx_fifo) { rc = -ENODEV; pr_err("%s: rx fifo not found\n", __func__); goto toc_init_fail; } einfo->tx_ch_desc->write_index = 0; einfo->rx_ch_desc->read_index = 0; rc = glink_core_register_transport(&einfo->xprt_if, &einfo->xprt_cfg); if (rc == -EPROBE_DEFER) goto reg_xprt_fail; if (rc) { pr_err("%s: glink core register transport failed: %d\n", __func__, rc); goto reg_xprt_fail; } einfo->irq_line = irq_line; rc = request_irq(irq_line, irq_handler, IRQF_TRIGGER_RISING | IRQF_NO_SUSPEND | IRQF_SHARED, node->name, einfo); if (rc < 0) { pr_err("%s: request_irq on %d failed: %d\n", __func__, irq_line, rc); goto request_irq_fail; } rc = enable_irq_wake(irq_line); if (rc < 0) pr_err("%s: enable_irq_wake() failed on %d\n", __func__, irq_line); register_debugfs_info(einfo); einfo->xprt_if.glink_core_if_ptr->link_up(&einfo->xprt_if); return 0; request_irq_fail: glink_core_unregister_transport(&einfo->xprt_if); reg_xprt_fail: toc_init_fail: flush_kthread_worker(&einfo->kworker); kthread_stop(einfo->task); einfo->task = NULL; kthread_fail: iounmap(msgram); msgram_ioremap_fail: iounmap(einfo->out_irq_reg); irq_ioremap_fail: mutex_lock(&probe_lock); edge_infos[einfo->remote_proc_id] = NULL; mutex_unlock(&probe_lock); invalid_key: missing_key: kfree(einfo); edge_info_alloc_fail: return rc; } #if defined(CONFIG_DEBUG_FS) /** * debug_edge() - generates formatted text output displaying current edge state * @s: File to send the output to. */ static void debug_edge(struct seq_file *s) { struct edge_info *einfo; struct glink_dbgfs_data *dfs_d; dfs_d = s->private; einfo = dfs_d->priv_data; /* * formatted, human readable edge state output, ie: * TX/RX fifo information: ID|EDGE |TX READ |TX WRITE |TX SIZE |RX READ |RX WRITE |RX SIZE ------------------------------------------------------------------------------- 01|mpss |0x00000128|0x00000128|0x00000800|0x00000256|0x00000256|0x00001000 * * Interrupt information: * EDGE |TX INT |RX INT * -------------------------------- * mpss |0x00000006|0x00000008 */ seq_puts(s, "TX/RX fifo information:\n"); seq_printf(s, "%2s|%-10s|%-10s|%-10s|%-10s|%-10s|%-10s|%-10s\n", "ID", "EDGE", "TX READ", "TX WRITE", "TX SIZE", "RX READ", "RX WRITE", "RX SIZE"); seq_puts(s, "-------------------------------------------------------------------------------\n"); if (!einfo) return; seq_printf(s, "%02i|%-10s|", einfo->remote_proc_id, einfo->xprt_cfg.edge); if (!einfo->rx_fifo) seq_puts(s, "Link Not Up\n"); else seq_printf(s, "0x%08X|0x%08X|0x%08X|0x%08X|0x%08X|0x%08X\n", einfo->tx_ch_desc->read_index, einfo->tx_ch_desc->write_index, einfo->tx_fifo_size, einfo->rx_ch_desc->read_index, einfo->rx_ch_desc->write_index, einfo->rx_fifo_size); seq_puts(s, "\nInterrupt information:\n"); seq_printf(s, "%-10s|%-10s|%-10s\n", "EDGE", "TX INT", "RX INT"); seq_puts(s, "--------------------------------\n"); seq_printf(s, "%-10s|0x%08X|0x%08X\n", einfo->xprt_cfg.edge, einfo->tx_irq_count, einfo->rx_irq_count); } /** * register_debugfs_info() - initialize debugfs device entries * @einfo: Pointer to specific edge_info for which register is called. */ static void register_debugfs_info(struct edge_info *einfo) { struct glink_dbgfs dfs; char *curr_dir_name; int dir_name_len; dir_name_len = strlen(einfo->xprt_cfg.edge) + strlen(einfo->xprt_cfg.name) + 2; curr_dir_name = kmalloc(dir_name_len, GFP_KERNEL); if (!curr_dir_name) { GLINK_ERR("%s: Memory allocation failed\n", __func__); return; } snprintf(curr_dir_name, dir_name_len, "%s_%s", einfo->xprt_cfg.edge, einfo->xprt_cfg.name); dfs.curr_name = curr_dir_name; dfs.par_name = "xprt"; dfs.b_dir_create = false; glink_debugfs_create("XPRT_INFO", debug_edge, &dfs, einfo, false); kfree(curr_dir_name); } #else static void register_debugfs_info(struct edge_info *einfo) { } #endif /* CONFIG_DEBUG_FS */ static struct of_device_id smem_match_table[] = { { .compatible = "qcom,glink-smem-native-xprt" }, {}, }; static struct platform_driver glink_smem_native_driver = { .probe = glink_smem_native_probe, .driver = { .name = "msm_glink_smem_native_xprt", .owner = THIS_MODULE, .of_match_table = smem_match_table, }, }; static struct of_device_id rpm_match_table[] = { { .compatible = "qcom,glink-rpm-native-xprt" }, {}, }; static struct platform_driver glink_rpm_native_driver = { .probe = glink_rpm_native_probe, .driver = { .name = "msm_glink_rpm_native_xprt", .owner = THIS_MODULE, .of_match_table = rpm_match_table, }, }; /** * ssr() - process a subsystem restart notification of a transport * @if_ptr: The transport to restart * * Return: 0 on success or standard Linux error code. */ static int ssr(struct glink_transport_if *if_ptr) { struct edge_info *einfo; struct deferred_cmd *cmd; struct device_node *node; uint32_t rpm_id; int rc = 0; einfo = container_of(if_ptr, struct edge_info, xprt_if); node = of_find_compatible_node(NULL, NULL, (*rpm_match_table).compatible); if (node) { rc = of_property_read_u32(node, "qcom,subsys-id", &rpm_id); if (rc) pr_err("%s: missing key\n", __func__); BUG_ON(einfo->remote_proc_id == rpm_id); } einfo->in_ssr = true; wake_up_all(&einfo->tx_blocked_queue); synchronize_srcu(&einfo->use_ref); while (!list_empty(&einfo->deferred_cmds)) { cmd = list_first_entry(&einfo->deferred_cmds, struct deferred_cmd, list_node); list_del(&cmd->list_node); kfree(cmd->data); kfree(cmd); } einfo->tx_resume_needed = false; einfo->tx_blocked_signal_sent = false; einfo->rx_fifo = NULL; einfo->rx_fifo_size = 0; einfo->tx_ch_desc->write_index = 0; einfo->rx_ch_desc->read_index = 0; einfo->xprt_if.glink_core_if_ptr->link_down(&einfo->xprt_if); return 0; } static int __init glink_smem_native_xprt_init(void) { int rc; rc = platform_driver_register(&glink_smem_native_driver); if (rc) { pr_err("%s: glink_smem_native_driver register failed %d\n", __func__, rc); return rc; } rc = platform_driver_register(&glink_rpm_native_driver); if (rc) { pr_err("%s: glink_rpm_native_driver register failed %d\n", __func__, rc); return rc; } return 0; } arch_initcall(glink_smem_native_xprt_init); MODULE_DESCRIPTION("MSM G-Link SMEM Native Transport"); MODULE_LICENSE("GPL v2");