blob: fa9546984c39ab2e14f2a20e64db3451d39f6a71 [file] [log] [blame]
/* Copyright (c) 2011-2014, 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.
*/
#define DEBUG
#include <linux/slab.h>
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/string.h>
#include <linux/errno.h>
#include <linux/init.h>
#include <linux/types.h>
#include <linux/delay.h>
#include <linux/err.h>
#include <linux/sched.h>
#include <linux/poll.h>
#include <linux/wakelock.h>
#include <linux/platform_device.h>
#include <linux/uaccess.h>
#include <linux/debugfs.h>
#include <linux/rwsem.h>
#include <asm/uaccess.h>
#include <asm/byteorder.h>
#include <mach/smem_log.h>
#include <mach/subsystem_notif.h>
#include <mach/msm_ipc_router.h>
#include <mach/msm_ipc_logging.h>
#include "ipc_router.h"
#include "modem_notifier.h"
#include "msm_ipc_router_security.h"
enum {
SMEM_LOG = 1U << 0,
RTR_DBG = 1U << 1,
R2R_MSG = 1U << 2,
R2R_RAW = 1U << 3,
NTFY_MSG = 1U << 4,
R2R_RAW_HDR = 1U << 5,
};
static int msm_ipc_router_debug_mask;
module_param_named(debug_mask, msm_ipc_router_debug_mask,
int, S_IRUGO | S_IWUSR | S_IWGRP);
static void *ipc_rtr_log_ctxt;
#define IPC_RTR_LOG_PAGES 5
#define DIAG(x...) pr_info("[RR] ERROR " x)
#if defined(DEBUG)
#define D(x...) do { \
if (ipc_rtr_log_ctxt) \
ipc_log_string(ipc_rtr_log_ctxt, x); \
if (msm_ipc_router_debug_mask & RTR_DBG) \
pr_info(x); \
} while (0)
#define RR(x...) do { \
if (ipc_rtr_log_ctxt) \
ipc_log_string(ipc_rtr_log_ctxt, x); \
if (msm_ipc_router_debug_mask & R2R_MSG) \
pr_info("[RR] "x); \
} while (0)
#define RAW(x...) do { \
if (msm_ipc_router_debug_mask & R2R_RAW) \
pr_info("[RAW] "x); \
} while (0)
#define NTFY(x...) do { \
if (msm_ipc_router_debug_mask & NTFY_MSG) \
pr_info("[NOTIFY] "x); \
} while (0)
#define RAW_HDR(x...) do { \
if (msm_ipc_router_debug_mask & R2R_RAW_HDR) \
pr_info("[HDR] "x); \
} while (0)
#else
#define D(x...) do { } while (0)
#define RR(x...) do { } while (0)
#define RAW(x...) do { } while (0)
#define RAW_HDR(x...) do { } while (0)
#define NTFY(x...) do { } while (0)
#endif
#define IPC_ROUTER_LOG_EVENT_ERROR 0x00
#define IPC_ROUTER_LOG_EVENT_TX 0x01
#define IPC_ROUTER_LOG_EVENT_RX 0x02
#define IPC_ROUTER_DUMMY_DEST_NODE 0xFFFFFFFF
static LIST_HEAD(control_ports);
static DECLARE_RWSEM(control_ports_lock_lha5);
#define LP_HASH_SIZE 32
static struct list_head local_ports[LP_HASH_SIZE];
static DECLARE_RWSEM(local_ports_lock_lha2);
/*
* Server info is organized as a hash table. The server's service ID is
* used to index into the hash table. The instance ID of most of the servers
* are 1 or 2. The service IDs are well distributed compared to the instance
* IDs and hence choosing service ID to index into this hash table optimizes
* the hash table operations like add, lookup, destroy.
*/
#define SRV_HASH_SIZE 32
static struct list_head server_list[SRV_HASH_SIZE];
static DECLARE_RWSEM(server_list_lock_lha2);
struct msm_ipc_server {
struct list_head list;
struct msm_ipc_port_name name;
char pdev_name[32];
int next_pdev_id;
int synced_sec_rule;
struct list_head server_port_list;
};
struct msm_ipc_server_port {
struct list_head list;
struct platform_device pdev;
struct msm_ipc_port_addr server_addr;
struct msm_ipc_router_xprt_info *xprt_info;
};
struct msm_ipc_resume_tx_port {
struct list_head list;
uint32_t port_id;
uint32_t node_id;
};
#define RP_HASH_SIZE 32
struct msm_ipc_router_remote_port {
struct list_head list;
uint32_t node_id;
uint32_t port_id;
uint32_t tx_quota_cnt;
struct mutex quota_lock_lhb2;
struct list_head resume_tx_port_list;
void *sec_rule;
struct msm_ipc_server *server;
};
struct msm_ipc_router_xprt_info {
struct list_head list;
struct msm_ipc_router_xprt *xprt;
uint32_t remote_node_id;
uint32_t initialized;
struct list_head pkt_list;
struct wake_lock wakelock;
struct mutex rx_lock_lhb2;
struct mutex tx_lock_lhb2;
uint32_t need_len;
uint32_t abort_data_read;
struct work_struct read_data;
struct workqueue_struct *workqueue;
};
#define RT_HASH_SIZE 4
struct msm_ipc_routing_table_entry {
struct list_head list;
uint32_t node_id;
uint32_t neighbor_node_id;
struct list_head remote_port_list[RP_HASH_SIZE];
struct msm_ipc_router_xprt_info *xprt_info;
struct rw_semaphore lock_lha4;
unsigned long num_tx_bytes;
unsigned long num_rx_bytes;
};
static struct list_head routing_table[RT_HASH_SIZE];
static DECLARE_RWSEM(routing_table_lock_lha3);
static int routing_table_inited;
static void do_read_data(struct work_struct *work);
static LIST_HEAD(xprt_info_list);
static DECLARE_RWSEM(xprt_info_list_lock_lha5);
static DECLARE_COMPLETION(msm_ipc_local_router_up);
#define IPC_ROUTER_INIT_TIMEOUT (10 * HZ)
static uint32_t next_port_id;
static DEFINE_MUTEX(next_port_id_lock_lha1);
static struct workqueue_struct *msm_ipc_router_workqueue;
enum {
DOWN,
UP,
};
static void init_routing_table(void)
{
int i;
for (i = 0; i < RT_HASH_SIZE; i++)
INIT_LIST_HEAD(&routing_table[i]);
}
static struct msm_ipc_routing_table_entry *alloc_routing_table_entry(
uint32_t node_id)
{
int i;
struct msm_ipc_routing_table_entry *rt_entry;
rt_entry = kmalloc(sizeof(struct msm_ipc_routing_table_entry),
GFP_KERNEL);
if (!rt_entry) {
pr_err("%s: rt_entry allocation failed for %d\n",
__func__, node_id);
return NULL;
}
for (i = 0; i < RP_HASH_SIZE; i++)
INIT_LIST_HEAD(&rt_entry->remote_port_list[i]);
init_rwsem(&rt_entry->lock_lha4);
rt_entry->node_id = node_id;
rt_entry->xprt_info = NULL;
return rt_entry;
}
/* Must be called with routing_table_lock_lha3 locked. */
static int add_routing_table_entry(
struct msm_ipc_routing_table_entry *rt_entry)
{
uint32_t key;
if (!rt_entry)
return -EINVAL;
key = (rt_entry->node_id % RT_HASH_SIZE);
list_add_tail(&rt_entry->list, &routing_table[key]);
return 0;
}
/* Must be called with routing_table_lock_lha3 locked. */
static struct msm_ipc_routing_table_entry *lookup_routing_table(
uint32_t node_id)
{
uint32_t key = (node_id % RT_HASH_SIZE);
struct msm_ipc_routing_table_entry *rt_entry;
list_for_each_entry(rt_entry, &routing_table[key], list) {
if (rt_entry->node_id == node_id)
return rt_entry;
}
return NULL;
}
struct rr_packet *rr_read(struct msm_ipc_router_xprt_info *xprt_info)
{
struct rr_packet *temp_pkt;
if (!xprt_info)
return NULL;
mutex_lock(&xprt_info->rx_lock_lhb2);
if (xprt_info->abort_data_read) {
mutex_unlock(&xprt_info->rx_lock_lhb2);
pr_err("%s detected SSR & exiting now\n",
xprt_info->xprt->name);
return NULL;
}
if (list_empty(&xprt_info->pkt_list)) {
mutex_unlock(&xprt_info->rx_lock_lhb2);
return NULL;
}
temp_pkt = list_first_entry(&xprt_info->pkt_list,
struct rr_packet, list);
list_del(&temp_pkt->list);
if (list_empty(&xprt_info->pkt_list))
wake_unlock(&xprt_info->wakelock);
mutex_unlock(&xprt_info->rx_lock_lhb2);
return temp_pkt;
}
struct rr_packet *clone_pkt(struct rr_packet *pkt)
{
struct rr_packet *cloned_pkt;
struct sk_buff *temp_skb, *cloned_skb;
struct sk_buff_head *pkt_fragment_q;
cloned_pkt = kzalloc(sizeof(struct rr_packet), GFP_KERNEL);
if (!cloned_pkt) {
pr_err("%s: failure\n", __func__);
return NULL;
}
memcpy(&(cloned_pkt->hdr), &(pkt->hdr), sizeof(struct rr_header_v1));
/* TODO: Copy optional headers, if available */
pkt_fragment_q = kmalloc(sizeof(struct sk_buff_head), GFP_KERNEL);
if (!pkt_fragment_q) {
pr_err("%s: pkt_frag_q alloc failure\n", __func__);
kfree(cloned_pkt);
return NULL;
}
skb_queue_head_init(pkt_fragment_q);
skb_queue_walk(pkt->pkt_fragment_q, temp_skb) {
cloned_skb = skb_clone(temp_skb, GFP_KERNEL);
if (!cloned_skb)
goto fail_clone;
skb_queue_tail(pkt_fragment_q, cloned_skb);
}
cloned_pkt->pkt_fragment_q = pkt_fragment_q;
cloned_pkt->length = pkt->length;
return cloned_pkt;
fail_clone:
while (!skb_queue_empty(pkt_fragment_q)) {
temp_skb = skb_dequeue(pkt_fragment_q);
kfree_skb(temp_skb);
}
kfree(pkt_fragment_q);
/* TODO: Free optional headers, if present */
kfree(cloned_pkt);
return NULL;
}
struct rr_packet *create_pkt(struct sk_buff_head *data)
{
struct rr_packet *pkt;
struct sk_buff *temp_skb;
pkt = kzalloc(sizeof(struct rr_packet), GFP_KERNEL);
if (!pkt) {
pr_err("%s: failure\n", __func__);
return NULL;
}
pkt->pkt_fragment_q = data;
skb_queue_walk(pkt->pkt_fragment_q, temp_skb)
pkt->length += temp_skb->len;
return pkt;
}
void release_pkt(struct rr_packet *pkt)
{
struct sk_buff *temp_skb;
if (!pkt)
return;
if (!pkt->pkt_fragment_q) {
kfree(pkt);
return;
}
while (!skb_queue_empty(pkt->pkt_fragment_q)) {
temp_skb = skb_dequeue(pkt->pkt_fragment_q);
kfree_skb(temp_skb);
}
kfree(pkt->pkt_fragment_q);
/* TODO: Free Optional headers, if present */
kfree(pkt);
return;
}
static struct sk_buff_head *msm_ipc_router_buf_to_skb(void *buf,
unsigned int buf_len)
{
struct sk_buff_head *skb_head;
struct sk_buff *skb;
int first = 1, offset = 0;
int skb_size, data_size;
void *data;
int last = 1;
int align_size;
skb_head = kmalloc(sizeof(struct sk_buff_head), GFP_KERNEL);
if (!skb_head) {
pr_err("%s: Couldnot allocate skb_head\n", __func__);
return NULL;
}
skb_queue_head_init(skb_head);
data_size = buf_len;
align_size = ALIGN_SIZE(data_size);
while (offset != buf_len) {
skb_size = data_size;
if (first)
skb_size += IPC_ROUTER_HDR_SIZE;
if (last)
skb_size += align_size;
skb = alloc_skb(skb_size, GFP_KERNEL);
if (!skb) {
if (skb_size <= (PAGE_SIZE/2)) {
pr_err("%s: cannot allocate skb\n", __func__);
goto buf_to_skb_error;
}
data_size = data_size / 2;
last = 0;
continue;
}
if (first) {
skb_reserve(skb, IPC_ROUTER_HDR_SIZE);
first = 0;
}
data = skb_put(skb, data_size);
memcpy(skb->data, buf + offset, data_size);
skb_queue_tail(skb_head, skb);
offset += data_size;
data_size = buf_len - offset;
last = 1;
}
return skb_head;
buf_to_skb_error:
while (!skb_queue_empty(skb_head)) {
skb = skb_dequeue(skb_head);
kfree_skb(skb);
}
kfree(skb_head);
return NULL;
}
static void *msm_ipc_router_skb_to_buf(struct sk_buff_head *skb_head,
unsigned int len)
{
struct sk_buff *temp;
unsigned int offset = 0, buf_len = 0, copy_len;
void *buf;
if (!skb_head) {
pr_err("%s: NULL skb_head\n", __func__);
return NULL;
}
temp = skb_peek(skb_head);
buf_len = len;
buf = kmalloc(buf_len, GFP_KERNEL);
if (!buf) {
pr_err("%s: cannot allocate buf\n", __func__);
return NULL;
}
skb_queue_walk(skb_head, temp) {
copy_len = buf_len < temp->len ? buf_len : temp->len;
memcpy(buf + offset, temp->data, copy_len);
offset += copy_len;
buf_len -= copy_len;
}
return buf;
}
void msm_ipc_router_free_skb(struct sk_buff_head *skb_head)
{
struct sk_buff *temp_skb;
if (!skb_head)
return;
while (!skb_queue_empty(skb_head)) {
temp_skb = skb_dequeue(skb_head);
kfree_skb(temp_skb);
}
kfree(skb_head);
}
/**
* extract_header_v1() - Extract IPC Router header of version 1
* @pkt: Packet structure into which the header has to be extraced.
* @skb: SKB from which the header has to be extracted.
*
* @return: 0 on success, standard Linux error codes on failure.
*/
static int extract_header_v1(struct rr_packet *pkt, struct sk_buff *skb)
{
if (!pkt || !skb) {
pr_err("%s: Invalid pkt or skb\n", __func__);
return -EINVAL;
}
memcpy(&pkt->hdr, skb->data, sizeof(struct rr_header_v1));
skb_pull(skb, sizeof(struct rr_header_v1));
pkt->length -= sizeof(struct rr_header_v1);
return 0;
}
/**
* extract_header_v2() - Extract IPC Router header of version 2
* @pkt: Packet structure into which the header has to be extraced.
* @skb: SKB from which the header has to be extracted.
*
* @return: 0 on success, standard Linux error codes on failure.
*/
static int extract_header_v2(struct rr_packet *pkt, struct sk_buff *skb)
{
struct rr_header_v2 *hdr;
if (!pkt || !skb) {
pr_err("%s: Invalid pkt or skb\n", __func__);
return -EINVAL;
}
hdr = (struct rr_header_v2 *)skb->data;
pkt->hdr.version = (uint32_t)hdr->version;
pkt->hdr.type = (uint32_t)hdr->type;
pkt->hdr.src_node_id = (uint32_t)hdr->src_node_id;
pkt->hdr.src_port_id = (uint32_t)hdr->src_port_id;
pkt->hdr.size = (uint32_t)hdr->size;
pkt->hdr.control_flag = (uint32_t)hdr->control_flag;
pkt->hdr.dst_node_id = (uint32_t)hdr->dst_node_id;
pkt->hdr.dst_port_id = (uint32_t)hdr->dst_port_id;
skb_pull(skb, sizeof(struct rr_header_v2));
pkt->length -= sizeof(struct rr_header_v2);
return 0;
}
/**
* extract_header() - Extract IPC Router header
* @pkt: Packet from which the header has to be extraced.
*
* @return: 0 on success, standard Linux error codes on failure.
*
* This function will check if the header version is v1 or v2 and invoke
* the corresponding helper function to extract the IPC Router header.
*/
static int extract_header(struct rr_packet *pkt)
{
struct sk_buff *temp_skb;
int ret;
if (!pkt) {
pr_err("%s: NULL PKT\n", __func__);
return -EINVAL;
}
temp_skb = skb_peek(pkt->pkt_fragment_q);
if (!temp_skb || !temp_skb->data) {
pr_err("%s: No SKBs in skb_queue\n", __func__);
return -EINVAL;
}
if (temp_skb->data[0] == IPC_ROUTER_V1) {
ret = extract_header_v1(pkt, temp_skb);
} else if (temp_skb->data[0] == IPC_ROUTER_V2) {
ret = extract_header_v2(pkt, temp_skb);
/* TODO: Extract optional headers if present */
} else {
pr_err("%s: Invalid Header version %02x\n",
__func__, temp_skb->data[0]);
print_hex_dump(KERN_ERR, "Header: ", DUMP_PREFIX_ADDRESS,
16, 1, temp_skb->data, pkt->length, true);
return -EINVAL;
}
return ret;
}
/**
* calc_tx_header_size() - Calculate header size to be reserved in SKB
* @pkt: Packet in which the space for header has to be reserved.
* @dst_xprt_info: XPRT through which the destination is reachable.
*
* @return: required header size on success,
* starndard Linux error codes on failure.
*
* This function is used to calculate the header size that has to be reserved
* in a transmit SKB. The header size is calculated based on the XPRT through
* which the destination node is reachable.
*/
static int calc_tx_header_size(struct rr_packet *pkt,
struct msm_ipc_router_xprt_info *dst_xprt_info)
{
int hdr_size = 0;
int xprt_version = 0;
struct msm_ipc_routing_table_entry *rt_entry;
struct msm_ipc_router_xprt_info *xprt_info = dst_xprt_info;
if (!pkt) {
pr_err("%s: NULL PKT\n", __func__);
return -EINVAL;
}
if (!xprt_info) {
rt_entry = lookup_routing_table(pkt->hdr.dst_node_id);
if (!rt_entry || !(rt_entry->xprt_info)) {
pr_err("%s: Node %d is not up\n",
__func__, pkt->hdr.dst_node_id);
return -ENODEV;
}
xprt_info = rt_entry->xprt_info;
}
if (xprt_info)
xprt_version = xprt_info->xprt->get_version(xprt_info->xprt);
if (xprt_version == IPC_ROUTER_V1) {
pkt->hdr.version = IPC_ROUTER_V1;
hdr_size = sizeof(struct rr_header_v1);
} else if (xprt_version == IPC_ROUTER_V2) {
pkt->hdr.version = IPC_ROUTER_V2;
hdr_size = sizeof(struct rr_header_v2);
/* TODO: Calculate optional header length, if present */
} else {
pr_err("%s: Invalid xprt_version %d\n",
__func__, xprt_version);
hdr_size = -EINVAL;
}
return hdr_size;
}
/**
* prepend_header_v1() - Prepend IPC Router header of version 1
* @pkt: Packet structure which contains the header info to be prepended.
* @hdr_size: Size of the header
*
* @return: 0 on success, standard Linux error codes on failure.
*/
static int prepend_header_v1(struct rr_packet *pkt, int hdr_size)
{
struct sk_buff *temp_skb;
struct rr_header_v1 *hdr;
if (!pkt || hdr_size <= 0) {
pr_err("%s: Invalid input parameters\n", __func__);
return -EINVAL;
}
temp_skb = skb_peek(pkt->pkt_fragment_q);
if (!temp_skb || !temp_skb->data) {
pr_err("%s: No SKBs in skb_queue\n", __func__);
return -EINVAL;
}
if (skb_headroom(temp_skb) < hdr_size) {
temp_skb = alloc_skb(hdr_size, GFP_KERNEL);
if (!temp_skb) {
pr_err("%s: Could not allocate SKB of size %d\n",
__func__, hdr_size);
return -ENOMEM;
}
}
hdr = (struct rr_header_v1 *)skb_push(temp_skb, hdr_size);
memcpy(hdr, &pkt->hdr, hdr_size);
if (temp_skb != skb_peek(pkt->pkt_fragment_q))
skb_queue_head(pkt->pkt_fragment_q, temp_skb);
pkt->length += hdr_size;
return 0;
}
/**
* prepend_header_v2() - Prepend IPC Router header of version 2
* @pkt: Packet structure which contains the header info to be prepended.
* @hdr_size: Size of the header
*
* @return: 0 on success, standard Linux error codes on failure.
*/
static int prepend_header_v2(struct rr_packet *pkt, int hdr_size)
{
struct sk_buff *temp_skb;
struct rr_header_v2 *hdr;
if (!pkt || hdr_size <= 0) {
pr_err("%s: Invalid input parameters\n", __func__);
return -EINVAL;
}
temp_skb = skb_peek(pkt->pkt_fragment_q);
if (!temp_skb || !temp_skb->data) {
pr_err("%s: No SKBs in skb_queue\n", __func__);
return -EINVAL;
}
if (skb_headroom(temp_skb) < hdr_size) {
temp_skb = alloc_skb(hdr_size, GFP_KERNEL);
if (!temp_skb) {
pr_err("%s: Could not allocate SKB of size %d\n",
__func__, hdr_size);
return -ENOMEM;
}
}
hdr = (struct rr_header_v2 *)skb_push(temp_skb, hdr_size);
hdr->version = (uint8_t)pkt->hdr.version;
hdr->type = (uint8_t)pkt->hdr.type;
hdr->control_flag = (uint16_t)pkt->hdr.control_flag;
hdr->size = (uint32_t)pkt->hdr.size;
hdr->src_node_id = (uint16_t)pkt->hdr.src_node_id;
hdr->src_port_id = (uint16_t)pkt->hdr.src_port_id;
hdr->dst_node_id = (uint16_t)pkt->hdr.dst_node_id;
hdr->dst_port_id = (uint16_t)pkt->hdr.dst_port_id;
/* TODO: Add optional headers, if present */
if (temp_skb != skb_peek(pkt->pkt_fragment_q))
skb_queue_head(pkt->pkt_fragment_q, temp_skb);
pkt->length += hdr_size;
return 0;
}
/**
* prepend_header() - Prepend IPC Router header
* @pkt: Packet structure which contains the header info to be prepended.
* @xprt_info: XPRT through which the packet is transmitted.
*
* @return: 0 on success, standard Linux error codes on failure.
*
* This function prepends the header to the packet to be transmitted. The
* IPC Router header version to be prepended depends on the XPRT through
* which the destination is reachable.
*/
static int prepend_header(struct rr_packet *pkt,
struct msm_ipc_router_xprt_info *xprt_info)
{
int hdr_size;
struct sk_buff *temp_skb;
if (!pkt) {
pr_err("%s: NULL PKT\n", __func__);
return -EINVAL;
}
temp_skb = skb_peek(pkt->pkt_fragment_q);
if (!temp_skb || !temp_skb->data) {
pr_err("%s: No SKBs in skb_queue\n", __func__);
return -EINVAL;
}
hdr_size = calc_tx_header_size(pkt, xprt_info);
if (hdr_size <= 0)
return hdr_size;
if (pkt->hdr.version == IPC_ROUTER_V1)
return prepend_header_v1(pkt, hdr_size);
else if (pkt->hdr.version == IPC_ROUTER_V2)
return prepend_header_v2(pkt, hdr_size);
else
return -EINVAL;
}
/**
* defragment_pkt() - Defragment and linearize the packet
* @pkt: Packet to be linearized.
*
* @return: 0 on success, standard Linux error codes on failure.
*
* Some packets contain fragments of data over multiple SKBs. If an XPRT
* does not supported fragmented writes, linearize multiple SKBs into one
* single SKB.
*/
static int defragment_pkt(struct rr_packet *pkt)
{
struct sk_buff *dst_skb, *src_skb, *temp_skb;
int offset = 0, buf_len = 0, copy_len;
void *buf;
int align_size;
if (!pkt || pkt->length <= 0) {
pr_err("%s: Invalid PKT\n", __func__);
return -EINVAL;
}
if (skb_queue_len(pkt->pkt_fragment_q) == 1)
return 0;
align_size = ALIGN_SIZE(pkt->length);
dst_skb = alloc_skb(pkt->length + align_size, GFP_KERNEL);
if (!dst_skb) {
pr_err("%s: could not allocate one skb of size %d\n",
__func__, pkt->length);
return -ENOMEM;
}
buf = skb_put(dst_skb, pkt->length);
buf_len = pkt->length;
skb_queue_walk(pkt->pkt_fragment_q, src_skb) {
copy_len = buf_len < src_skb->len ? buf_len : src_skb->len;
memcpy(buf + offset, src_skb->data, copy_len);
offset += copy_len;
buf_len -= copy_len;
}
while (!skb_queue_empty(pkt->pkt_fragment_q)) {
temp_skb = skb_dequeue(pkt->pkt_fragment_q);
kfree_skb(temp_skb);
}
skb_queue_tail(pkt->pkt_fragment_q, dst_skb);
return 0;
}
static int post_pkt_to_port(struct msm_ipc_port *port_ptr,
struct rr_packet *pkt, int clone)
{
struct rr_packet *temp_pkt = pkt;
void (*notify)(unsigned event, void *priv);
if (unlikely(!port_ptr || !pkt))
return -EINVAL;
if (clone) {
temp_pkt = clone_pkt(pkt);
if (!temp_pkt) {
pr_err("%s: Error cloning packet for port %08x:%08x\n",
__func__, port_ptr->this_port.node_id,
port_ptr->this_port.port_id);
return -ENOMEM;
}
}
mutex_lock(&port_ptr->port_rx_q_lock_lhb3);
wake_lock(&port_ptr->port_rx_wake_lock);
list_add_tail(&temp_pkt->list, &port_ptr->port_rx_q);
wake_up(&port_ptr->port_rx_wait_q);
notify = port_ptr->notify;
mutex_unlock(&port_ptr->port_rx_q_lock_lhb3);
if (notify)
notify(MSM_IPC_ROUTER_READ_CB, port_ptr->priv);
return 0;
}
static int post_control_ports(struct rr_packet *pkt)
{
struct msm_ipc_port *port_ptr;
if (!pkt)
return -EINVAL;
down_read(&control_ports_lock_lha5);
list_for_each_entry(port_ptr, &control_ports, list)
post_pkt_to_port(port_ptr, pkt, 1);
up_read(&control_ports_lock_lha5);
return 0;
}
static uint32_t allocate_port_id(void)
{
uint32_t port_id = 0, prev_port_id, key;
struct msm_ipc_port *port_ptr;
mutex_lock(&next_port_id_lock_lha1);
prev_port_id = next_port_id;
down_read(&local_ports_lock_lha2);
do {
next_port_id++;
if ((next_port_id & IPC_ROUTER_ADDRESS) == IPC_ROUTER_ADDRESS)
next_port_id = 1;
key = (next_port_id & (LP_HASH_SIZE - 1));
if (list_empty(&local_ports[key])) {
port_id = next_port_id;
break;
}
list_for_each_entry(port_ptr, &local_ports[key], list) {
if (port_ptr->this_port.port_id == next_port_id) {
port_id = next_port_id;
break;
}
}
if (!port_id) {
port_id = next_port_id;
break;
}
port_id = 0;
} while (next_port_id != prev_port_id);
up_read(&local_ports_lock_lha2);
mutex_unlock(&next_port_id_lock_lha1);
return port_id;
}
void msm_ipc_router_add_local_port(struct msm_ipc_port *port_ptr)
{
uint32_t key;
if (!port_ptr)
return;
key = (port_ptr->this_port.port_id & (LP_HASH_SIZE - 1));
down_write(&local_ports_lock_lha2);
list_add_tail(&port_ptr->list, &local_ports[key]);
up_write(&local_ports_lock_lha2);
}
struct msm_ipc_port *msm_ipc_router_create_raw_port(void *endpoint,
void (*notify)(unsigned event, void *priv),
void *priv)
{
struct msm_ipc_port *port_ptr;
port_ptr = kzalloc(sizeof(struct msm_ipc_port), GFP_KERNEL);
if (!port_ptr)
return NULL;
port_ptr->this_port.node_id = IPC_ROUTER_NID_LOCAL;
port_ptr->this_port.port_id = allocate_port_id();
if (!port_ptr->this_port.port_id) {
pr_err("%s: All port ids are in use\n", __func__);
kfree(port_ptr);
return NULL;
}
spin_lock_init(&port_ptr->port_lock);
INIT_LIST_HEAD(&port_ptr->port_rx_q);
mutex_init(&port_ptr->port_rx_q_lock_lhb3);
init_waitqueue_head(&port_ptr->port_rx_wait_q);
snprintf(port_ptr->rx_wakelock_name, MAX_WAKELOCK_NAME_SZ,
"ipc%08x_%s",
port_ptr->this_port.port_id,
current->comm);
wake_lock_init(&port_ptr->port_rx_wake_lock,
WAKE_LOCK_SUSPEND, port_ptr->rx_wakelock_name);
port_ptr->endpoint = endpoint;
port_ptr->notify = notify;
port_ptr->priv = priv;
msm_ipc_router_add_local_port(port_ptr);
return port_ptr;
}
/* Must be called with local_ports_lock_lha2 locked. */
static struct msm_ipc_port *msm_ipc_router_lookup_local_port(uint32_t port_id)
{
int key = (port_id & (LP_HASH_SIZE - 1));
struct msm_ipc_port *port_ptr;
list_for_each_entry(port_ptr, &local_ports[key], list) {
if (port_ptr->this_port.port_id == port_id) {
return port_ptr;
}
}
return NULL;
}
/* Must be called with routing_table_lock_lha3 locked. */
static struct msm_ipc_router_remote_port *msm_ipc_router_lookup_remote_port(
uint32_t node_id,
uint32_t port_id)
{
struct msm_ipc_router_remote_port *rport_ptr;
struct msm_ipc_routing_table_entry *rt_entry;
int key = (port_id & (RP_HASH_SIZE - 1));
rt_entry = lookup_routing_table(node_id);
if (!rt_entry) {
pr_err("%s: Node is not up\n", __func__);
return NULL;
}
down_read(&rt_entry->lock_lha4);
list_for_each_entry(rport_ptr,
&rt_entry->remote_port_list[key], list) {
if (rport_ptr->port_id == port_id) {
up_read(&rt_entry->lock_lha4);
return rport_ptr;
}
}
up_read(&rt_entry->lock_lha4);
return NULL;
}
/* Must be called with routing_table_lock_lha3 locked. */
static struct msm_ipc_router_remote_port *msm_ipc_router_create_remote_port(
uint32_t node_id,
uint32_t port_id)
{
struct msm_ipc_router_remote_port *rport_ptr;
struct msm_ipc_routing_table_entry *rt_entry;
int key = (port_id & (RP_HASH_SIZE - 1));
rt_entry = lookup_routing_table(node_id);
if (!rt_entry) {
pr_err("%s: Node is not up\n", __func__);
return NULL;
}
rport_ptr = kmalloc(sizeof(struct msm_ipc_router_remote_port),
GFP_KERNEL);
if (!rport_ptr) {
pr_err("%s: Remote port alloc failed\n", __func__);
return NULL;
}
rport_ptr->port_id = port_id;
rport_ptr->node_id = node_id;
rport_ptr->sec_rule = NULL;
rport_ptr->server = NULL;
rport_ptr->tx_quota_cnt = 0;
mutex_init(&rport_ptr->quota_lock_lhb2);
INIT_LIST_HEAD(&rport_ptr->resume_tx_port_list);
down_write(&rt_entry->lock_lha4);
list_add_tail(&rport_ptr->list,
&rt_entry->remote_port_list[key]);
up_write(&rt_entry->lock_lha4);
return rport_ptr;
}
/**
* msm_ipc_router_free_resume_tx_port() - Free the resume_tx ports
* @rport_ptr: Pointer to the remote port.
*
* This function deletes all the resume_tx ports associated with a remote port
* and frees the memory allocated to each resume_tx port.
*
* Must be called with rport_ptr->quota_lock_lhb2 locked.
*/
static void msm_ipc_router_free_resume_tx_port(
struct msm_ipc_router_remote_port *rport_ptr)
{
struct msm_ipc_resume_tx_port *rtx_port, *tmp_rtx_port;
list_for_each_entry_safe(rtx_port, tmp_rtx_port,
&rport_ptr->resume_tx_port_list, list) {
list_del(&rtx_port->list);
kfree(rtx_port);
}
}
/**
* msm_ipc_router_lookup_resume_tx_port() - Lookup resume_tx port list
* @rport_ptr: Remote port whose resume_tx port list needs to be looked.
* @port_id: Port ID which needs to be looked from the list.
*
* return 1 if the port_id is found in the list, else 0.
*
* This function is used to lookup the existence of a local port in
* remote port's resume_tx list. This function is used to ensure that
* the same port is not added to the remote_port's resume_tx list repeatedly.
*
* Must be called with rport_ptr->quota_lock_lhb2 locked.
*/
static int msm_ipc_router_lookup_resume_tx_port(
struct msm_ipc_router_remote_port *rport_ptr, uint32_t port_id)
{
struct msm_ipc_resume_tx_port *rtx_port;
list_for_each_entry(rtx_port, &rport_ptr->resume_tx_port_list, list) {
if (port_id == rtx_port->port_id)
return 1;
}
return 0;
}
/**
* post_resume_tx() - Post the resume_tx event
* @rport_ptr: Pointer to the remote port
* @pkt : The data packet that is received on a resume_tx event
*
* This function informs about the reception of the resume_tx message from a
* remote port pointed by rport_ptr to all the local ports that are in the
* resume_tx_ports_list of this remote port. On posting the information, this
* function sequentially deletes each entry in the resume_tx_port_list of the
* remote port.
*
* Must be called with rport_ptr->quota_lock_lhb2 locked.
*/
static void post_resume_tx(struct msm_ipc_router_remote_port *rport_ptr,
struct rr_packet *pkt)
{
struct msm_ipc_resume_tx_port *rtx_port, *tmp_rtx_port;
struct msm_ipc_port *local_port;
list_for_each_entry_safe(rtx_port, tmp_rtx_port,
&rport_ptr->resume_tx_port_list, list) {
local_port =
msm_ipc_router_lookup_local_port(rtx_port->port_id);
if (local_port && local_port->notify)
local_port->notify(MSM_IPC_ROUTER_RESUME_TX,
local_port->priv);
else if (local_port)
post_pkt_to_port(local_port, pkt, 1);
else
pr_err("%s: Local Port %d not Found",
__func__, rtx_port->port_id);
list_del(&rtx_port->list);
kfree(rtx_port);
}
}
/* Must be called with routing_table_lock_lha3 locked. */
static void msm_ipc_router_destroy_remote_port(
struct msm_ipc_router_remote_port *rport_ptr)
{
uint32_t node_id;
struct msm_ipc_routing_table_entry *rt_entry;
if (!rport_ptr)
return;
node_id = rport_ptr->node_id;
rt_entry = lookup_routing_table(node_id);
if (!rt_entry) {
pr_err("%s: Node %d is not up\n", __func__, node_id);
return;
}
down_write(&rt_entry->lock_lha4);
list_del(&rport_ptr->list);
up_write(&rt_entry->lock_lha4);
mutex_lock(&rport_ptr->quota_lock_lhb2);
msm_ipc_router_free_resume_tx_port(rport_ptr);
mutex_unlock(&rport_ptr->quota_lock_lhb2);
kfree(rport_ptr);
return;
}
/**
* msm_ipc_router_lookup_server() - Lookup server information
* @service: Service ID of the server info to be looked up.
* @instance: Instance ID of the server info to be looked up.
* @node_id: Node/Processor ID in which the server is hosted.
* @port_id: Port ID within the node in which the server is hosted.
*
* @return: If found Pointer to server structure, else NULL.
*
* Note1: Lock the server_list_lock_lha2 before accessing this function.
* Note2: If the <node_id:port_id> are <0:0>, then the lookup is restricted
* to <service:instance>. Used only when a client wants to send a
* message to any QMI server.
*/
static struct msm_ipc_server *msm_ipc_router_lookup_server(
uint32_t service,
uint32_t instance,
uint32_t node_id,
uint32_t port_id)
{
struct msm_ipc_server *server;
struct msm_ipc_server_port *server_port;
int key = (service & (SRV_HASH_SIZE - 1));
list_for_each_entry(server, &server_list[key], list) {
if ((server->name.service != service) ||
(server->name.instance != instance))
continue;
if ((node_id == 0) && (port_id == 0))
return server;
list_for_each_entry(server_port, &server->server_port_list,
list) {
if ((server_port->server_addr.node_id == node_id) &&
(server_port->server_addr.port_id == port_id))
return server;
}
}
return NULL;
}
static void dummy_release(struct device *dev)
{
}
/**
* msm_ipc_router_create_server() - Add server info to hash table
* @service: Service ID of the server info to be created.
* @instance: Instance ID of the server info to be created.
* @node_id: Node/Processor ID in which the server is hosted.
* @port_id: Port ID within the node in which the server is hosted.
* @xprt_info: XPRT through which the node hosting the server is reached.
*
* @return: Pointer to server structure on success, else NULL.
*
* This function adds the server info to the hash table. If the same
* server(i.e. <service_id:instance_id>) is hosted in different nodes,
* they are maintained as list of "server_port" under "server" structure.
* Note: Lock the server_list_lock_lha2 before accessing this function.
*/
static struct msm_ipc_server *msm_ipc_router_create_server(
uint32_t service,
uint32_t instance,
uint32_t node_id,
uint32_t port_id,
struct msm_ipc_router_xprt_info *xprt_info)
{
struct msm_ipc_server *server = NULL;
struct msm_ipc_server_port *server_port;
int key = (service & (SRV_HASH_SIZE - 1));
list_for_each_entry(server, &server_list[key], list) {
if ((server->name.service == service) &&
(server->name.instance == instance))
goto create_srv_port;
}
server = kzalloc(sizeof(struct msm_ipc_server), GFP_KERNEL);
if (!server) {
pr_err("%s: Server allocation failed\n", __func__);
return NULL;
}
server->name.service = service;
server->name.instance = instance;
server->synced_sec_rule = 0;
INIT_LIST_HEAD(&server->server_port_list);
list_add_tail(&server->list, &server_list[key]);
scnprintf(server->pdev_name, sizeof(server->pdev_name),
"QMI%08x:%08x", service, instance);
server->next_pdev_id = 1;
create_srv_port:
server_port = kzalloc(sizeof(struct msm_ipc_server_port), GFP_KERNEL);
if (!server_port) {
if (list_empty(&server->server_port_list)) {
list_del(&server->list);
kfree(server);
}
pr_err("%s: Server Port allocation failed\n", __func__);
return NULL;
}
server_port->server_addr.node_id = node_id;
server_port->server_addr.port_id = port_id;
server_port->xprt_info = xprt_info;
list_add_tail(&server_port->list, &server->server_port_list);
server_port->pdev.name = server->pdev_name;
server_port->pdev.id = server->next_pdev_id++;
server_port->pdev.dev.release = dummy_release;
platform_device_register(&server_port->pdev);
return server;
}
/**
* msm_ipc_router_destroy_server() - Remove server info from hash table
* @server: Server info to be removed.
* @node_id: Node/Processor ID in which the server is hosted.
* @port_id: Port ID within the node in which the server is hosted.
*
* This function removes the server_port identified using <node_id:port_id>
* from the server structure. If the server_port list under server structure
* is empty after removal, then remove the server structure from the server
* hash table.
* Note: Lock the server_list_lock_lha2 before accessing this function.
*/
static void msm_ipc_router_destroy_server(struct msm_ipc_server *server,
uint32_t node_id, uint32_t port_id)
{
struct msm_ipc_server_port *server_port;
if (!server)
return;
list_for_each_entry(server_port, &server->server_port_list, list) {
if ((server_port->server_addr.node_id == node_id) &&
(server_port->server_addr.port_id == port_id))
break;
}
if (server_port) {
platform_device_unregister(&server_port->pdev);
list_del(&server_port->list);
kfree(server_port);
}
if (list_empty(&server->server_port_list)) {
list_del(&server->list);
kfree(server);
}
return;
}
static int msm_ipc_router_send_control_msg(
struct msm_ipc_router_xprt_info *xprt_info,
union rr_control_msg *msg,
uint32_t dst_node_id)
{
struct rr_packet *pkt;
struct sk_buff *ipc_rtr_pkt;
struct rr_header_v1 *hdr;
int pkt_size;
void *data;
struct sk_buff_head *pkt_fragment_q;
int ret;
if (!xprt_info || ((msg->cmd != IPC_ROUTER_CTRL_CMD_HELLO) &&
!xprt_info->initialized)) {
pr_err("%s: xprt_info not initialized\n", __func__);
return -EINVAL;
}
if (xprt_info->remote_node_id == IPC_ROUTER_NID_LOCAL)
return 0;
pkt = kzalloc(sizeof(struct rr_packet), GFP_KERNEL);
if (!pkt) {
pr_err("%s: pkt alloc failed\n", __func__);
return -ENOMEM;
}
pkt_fragment_q = kmalloc(sizeof(struct sk_buff_head), GFP_KERNEL);
if (!pkt_fragment_q) {
pr_err("%s: pkt_fragment_q alloc failed\n", __func__);
kfree(pkt);
return -ENOMEM;
}
skb_queue_head_init(pkt_fragment_q);
pkt_size = IPC_ROUTER_HDR_SIZE + sizeof(*msg);
ipc_rtr_pkt = alloc_skb(pkt_size, GFP_KERNEL);
if (!ipc_rtr_pkt) {
pr_err("%s: ipc_rtr_pkt alloc failed\n", __func__);
kfree(pkt_fragment_q);
kfree(pkt);
return -ENOMEM;
}
skb_reserve(ipc_rtr_pkt, IPC_ROUTER_HDR_SIZE);
data = skb_put(ipc_rtr_pkt, sizeof(*msg));
memcpy(data, msg, sizeof(*msg));
skb_queue_tail(pkt_fragment_q, ipc_rtr_pkt);
pkt->pkt_fragment_q = pkt_fragment_q;
pkt->length = sizeof(*msg);
hdr = &(pkt->hdr);
hdr->version = IPC_ROUTER_V1;
hdr->type = msg->cmd;
hdr->src_node_id = IPC_ROUTER_NID_LOCAL;
hdr->src_port_id = IPC_ROUTER_ADDRESS;
hdr->control_flag = 0;
hdr->size = sizeof(*msg);
if (hdr->type == IPC_ROUTER_CTRL_CMD_RESUME_TX)
hdr->dst_node_id = dst_node_id;
else
hdr->dst_node_id = xprt_info->remote_node_id;
hdr->dst_port_id = IPC_ROUTER_ADDRESS;
mutex_lock(&xprt_info->tx_lock_lhb2);
ret = prepend_header(pkt, xprt_info);
if (ret < 0) {
mutex_unlock(&xprt_info->tx_lock_lhb2);
pr_err("%s: Prepend Header failed\n", __func__);
release_pkt(pkt);
return ret;
}
ret = xprt_info->xprt->write(pkt, pkt->length, xprt_info->xprt);
mutex_unlock(&xprt_info->tx_lock_lhb2);
release_pkt(pkt);
return ret;
}
static int msm_ipc_router_send_server_list(uint32_t node_id,
struct msm_ipc_router_xprt_info *xprt_info)
{
union rr_control_msg ctl;
struct msm_ipc_server *server;
struct msm_ipc_server_port *server_port;
int i;
if (!xprt_info || !xprt_info->initialized) {
pr_err("%s: Xprt info not initialized\n", __func__);
return -EINVAL;
}
memset(&ctl, 0, sizeof(ctl));
ctl.cmd = IPC_ROUTER_CTRL_CMD_NEW_SERVER;
for (i = 0; i < SRV_HASH_SIZE; i++) {
list_for_each_entry(server, &server_list[i], list) {
ctl.srv.service = server->name.service;
ctl.srv.instance = server->name.instance;
list_for_each_entry(server_port,
&server->server_port_list, list) {
if (server_port->server_addr.node_id !=
node_id)
continue;
ctl.srv.node_id =
server_port->server_addr.node_id;
ctl.srv.port_id =
server_port->server_addr.port_id;
msm_ipc_router_send_control_msg(xprt_info,
&ctl, IPC_ROUTER_DUMMY_DEST_NODE);
}
}
}
return 0;
}
#if defined(DEBUG)
static char *type_to_str(int i)
{
switch (i) {
case IPC_ROUTER_CTRL_CMD_DATA:
return "data ";
case IPC_ROUTER_CTRL_CMD_HELLO:
return "hello ";
case IPC_ROUTER_CTRL_CMD_BYE:
return "bye ";
case IPC_ROUTER_CTRL_CMD_NEW_SERVER:
return "new_srvr";
case IPC_ROUTER_CTRL_CMD_REMOVE_SERVER:
return "rmv_srvr";
case IPC_ROUTER_CTRL_CMD_REMOVE_CLIENT:
return "rmv_clnt";
case IPC_ROUTER_CTRL_CMD_RESUME_TX:
return "resum_tx";
case IPC_ROUTER_CTRL_CMD_EXIT:
return "cmd_exit";
default:
return "invalid";
}
}
#endif
static int broadcast_ctl_msg_locally(union rr_control_msg *msg)
{
struct rr_packet *pkt;
struct sk_buff *ipc_rtr_pkt;
struct rr_header_v1 *hdr;
int pkt_size;
void *data;
struct sk_buff_head *pkt_fragment_q;
int ret;
pkt = kzalloc(sizeof(struct rr_packet), GFP_KERNEL);
if (!pkt) {
pr_err("%s: pkt alloc failed\n", __func__);
return -ENOMEM;
}
pkt_fragment_q = kmalloc(sizeof(struct sk_buff_head), GFP_KERNEL);
if (!pkt_fragment_q) {
pr_err("%s: pkt_fragment_q alloc failed\n", __func__);
kfree(pkt);
return -ENOMEM;
}
skb_queue_head_init(pkt_fragment_q);
pkt_size = sizeof(*msg);
ipc_rtr_pkt = alloc_skb(pkt_size, GFP_KERNEL);
if (!ipc_rtr_pkt) {
pr_err("%s: ipc_rtr_pkt alloc failed\n", __func__);
kfree(pkt_fragment_q);
kfree(pkt);
return -ENOMEM;
}
data = skb_put(ipc_rtr_pkt, sizeof(*msg));
memcpy(data, msg, sizeof(*msg));
hdr = &(pkt->hdr);
hdr->version = IPC_ROUTER_V1;
hdr->type = msg->cmd;
hdr->src_node_id = IPC_ROUTER_NID_LOCAL;
hdr->src_port_id = IPC_ROUTER_ADDRESS;
hdr->control_flag = 0;
hdr->size = sizeof(*msg);
hdr->dst_node_id = IPC_ROUTER_NID_LOCAL;
hdr->dst_port_id = IPC_ROUTER_ADDRESS;
skb_queue_tail(pkt_fragment_q, ipc_rtr_pkt);
pkt->pkt_fragment_q = pkt_fragment_q;
pkt->length = pkt_size;
ret = post_control_ports(pkt);
release_pkt(pkt);
return ret;
}
static int broadcast_ctl_msg(union rr_control_msg *ctl)
{
struct msm_ipc_router_xprt_info *xprt_info;
down_read(&xprt_info_list_lock_lha5);
list_for_each_entry(xprt_info, &xprt_info_list, list) {
msm_ipc_router_send_control_msg(xprt_info, ctl,
IPC_ROUTER_DUMMY_DEST_NODE);
}
up_read(&xprt_info_list_lock_lha5);
return 0;
}
static int relay_ctl_msg(struct msm_ipc_router_xprt_info *xprt_info,
union rr_control_msg *ctl)
{
struct msm_ipc_router_xprt_info *fwd_xprt_info;
if (!xprt_info || !ctl)
return -EINVAL;
down_read(&xprt_info_list_lock_lha5);
list_for_each_entry(fwd_xprt_info, &xprt_info_list, list) {
if (xprt_info->xprt->link_id != fwd_xprt_info->xprt->link_id)
msm_ipc_router_send_control_msg(fwd_xprt_info, ctl,
IPC_ROUTER_DUMMY_DEST_NODE);
}
up_read(&xprt_info_list_lock_lha5);
return 0;
}
static int forward_msg(struct msm_ipc_router_xprt_info *xprt_info,
struct rr_packet *pkt)
{
struct rr_header_v1 *hdr;
struct msm_ipc_router_xprt_info *fwd_xprt_info;
struct msm_ipc_routing_table_entry *rt_entry;
int ret = 0;
int fwd_xprt_option;
if (!xprt_info || !pkt)
return -EINVAL;
hdr = &(pkt->hdr);
down_read(&routing_table_lock_lha3);
rt_entry = lookup_routing_table(hdr->dst_node_id);
if (!(rt_entry) || !(rt_entry->xprt_info)) {
pr_err("%s: Routing table not initialized\n", __func__);
ret = -ENODEV;
goto fm_error1;
}
down_read(&rt_entry->lock_lha4);
fwd_xprt_info = rt_entry->xprt_info;
ret = prepend_header(pkt, fwd_xprt_info);
if (ret < 0) {
pr_err("%s: Prepend Header failed\n", __func__);
goto fm_error2;
}
fwd_xprt_option = fwd_xprt_info->xprt->get_option(fwd_xprt_info->xprt);
if (!(fwd_xprt_option & FRAG_PKT_WRITE_ENABLE)) {
ret = defragment_pkt(pkt);
if (ret < 0)
goto fm_error2;
}
mutex_lock(&fwd_xprt_info->tx_lock_lhb2);
if (xprt_info->remote_node_id == fwd_xprt_info->remote_node_id) {
pr_err("%s: Discarding Command to route back\n", __func__);
ret = -EINVAL;
goto fm_error3;
}
if (xprt_info->xprt->link_id == fwd_xprt_info->xprt->link_id) {
pr_err("%s: DST in the same cluster\n", __func__);
ret = 0;
goto fm_error3;
}
fwd_xprt_info->xprt->write(pkt, pkt->length, fwd_xprt_info->xprt);
fm_error3:
mutex_unlock(&fwd_xprt_info->tx_lock_lhb2);
fm_error2:
up_read(&rt_entry->lock_lha4);
fm_error1:
up_read(&routing_table_lock_lha3);
return ret;
}
static int msm_ipc_router_send_remove_client(struct comm_mode_info *mode_info,
uint32_t node_id, uint32_t port_id)
{
union rr_control_msg msg;
struct msm_ipc_router_xprt_info *tmp_xprt_info;
int mode;
void *xprt_info;
int rc = 0;
if (!mode_info) {
pr_err("%s: NULL mode_info\n", __func__);
return -EINVAL;
}
mode = mode_info->mode;
xprt_info = mode_info->xprt_info;
memset(&msg, 0, sizeof(msg));
msg.cmd = IPC_ROUTER_CTRL_CMD_REMOVE_CLIENT;
msg.cli.node_id = node_id;
msg.cli.port_id = port_id;
if ((mode == SINGLE_LINK_MODE) && xprt_info) {
down_read(&xprt_info_list_lock_lha5);
list_for_each_entry(tmp_xprt_info, &xprt_info_list, list) {
if (tmp_xprt_info != xprt_info)
continue;
msm_ipc_router_send_control_msg(tmp_xprt_info, &msg,
IPC_ROUTER_DUMMY_DEST_NODE);
break;
}
up_read(&xprt_info_list_lock_lha5);
} else if ((mode == SINGLE_LINK_MODE) && !xprt_info) {
broadcast_ctl_msg_locally(&msg);
} else if (mode == MULTI_LINK_MODE) {
broadcast_ctl_msg(&msg);
broadcast_ctl_msg_locally(&msg);
} else if (mode != NULL_MODE) {
pr_err("%s: Invalid mode(%d) + xprt_inf(%p) for %08x:%08x\n",
__func__, mode, xprt_info, node_id, port_id);
rc = -EINVAL;
}
return rc;
}
static void update_comm_mode_info(struct comm_mode_info *mode_info,
struct msm_ipc_router_xprt_info *xprt_info)
{
if (!mode_info) {
pr_err("%s: NULL mode_info\n", __func__);
return;
}
if (mode_info->mode == NULL_MODE) {
mode_info->xprt_info = xprt_info;
mode_info->mode = SINGLE_LINK_MODE;
} else if (mode_info->mode == SINGLE_LINK_MODE &&
mode_info->xprt_info != xprt_info) {
mode_info->mode = MULTI_LINK_MODE;
}
return;
}
static void cleanup_rmt_server(struct msm_ipc_router_xprt_info *xprt_info,
struct msm_ipc_router_remote_port *rport_ptr)
{
union rr_control_msg ctl;
struct msm_ipc_server *server = rport_ptr->server;
D("Remove server %08x:%08x - %08x:%08x",
server->name.service, server->name.instance,
rport_ptr->node_id, rport_ptr->port_id);
memset(&ctl, 0, sizeof(ctl));
ctl.cmd = IPC_ROUTER_CTRL_CMD_REMOVE_SERVER;
ctl.srv.service = server->name.service;
ctl.srv.instance = server->name.instance;
ctl.srv.node_id = rport_ptr->node_id;
ctl.srv.port_id = rport_ptr->port_id;
relay_ctl_msg(xprt_info, &ctl);
broadcast_ctl_msg_locally(&ctl);
msm_ipc_router_destroy_server(server,
rport_ptr->node_id, rport_ptr->port_id);
}
static void cleanup_rmt_ports(struct msm_ipc_router_xprt_info *xprt_info,
struct msm_ipc_routing_table_entry *rt_entry)
{
struct msm_ipc_router_remote_port *rport_ptr, *tmp_rport_ptr;
union rr_control_msg ctl;
int j;
memset(&ctl, 0, sizeof(ctl));
for (j = 0; j < RP_HASH_SIZE; j++) {
list_for_each_entry_safe(rport_ptr, tmp_rport_ptr,
&rt_entry->remote_port_list[j], list) {
list_del(&rport_ptr->list);
mutex_lock(&rport_ptr->quota_lock_lhb2);
msm_ipc_router_free_resume_tx_port(rport_ptr);
mutex_unlock(&rport_ptr->quota_lock_lhb2);
if (rport_ptr->server)
cleanup_rmt_server(xprt_info, rport_ptr);
ctl.cmd = IPC_ROUTER_CTRL_CMD_REMOVE_CLIENT;
ctl.cli.node_id = rport_ptr->node_id;
ctl.cli.port_id = rport_ptr->port_id;
relay_ctl_msg(xprt_info, &ctl);
broadcast_ctl_msg_locally(&ctl);
kfree(rport_ptr);
}
}
}
static void msm_ipc_cleanup_routing_table(
struct msm_ipc_router_xprt_info *xprt_info)
{
int i;
struct msm_ipc_routing_table_entry *rt_entry, *tmp_rt_entry;
if (!xprt_info) {
pr_err("%s: Invalid xprt_info\n", __func__);
return;
}
down_write(&server_list_lock_lha2);
down_write(&routing_table_lock_lha3);
for (i = 0; i < RT_HASH_SIZE; i++) {
list_for_each_entry_safe(rt_entry, tmp_rt_entry,
&routing_table[i], list) {
down_write(&rt_entry->lock_lha4);
if (rt_entry->xprt_info != xprt_info) {
up_write(&rt_entry->lock_lha4);
continue;
}
cleanup_rmt_ports(xprt_info, rt_entry);
rt_entry->xprt_info = NULL;
up_write(&rt_entry->lock_lha4);
list_del(&rt_entry->list);
kfree(rt_entry);
}
}
up_write(&routing_table_lock_lha3);
up_write(&server_list_lock_lha2);
}
/**
* sync_sec_rule() - Synchrnoize the security rule into the server structure
* @server: Server structure where the rule has to be synchronized.
* @rule: Security tule to be synchronized.
*
* This function is used to update the server structure with the security
* rule configured for the <service:instance> corresponding to that server.
*/
static void sync_sec_rule(struct msm_ipc_server *server, void *rule)
{
struct msm_ipc_server_port *server_port;
struct msm_ipc_router_remote_port *rport_ptr = NULL;
down_read(&routing_table_lock_lha3);
list_for_each_entry(server_port, &server->server_port_list, list) {
rport_ptr = msm_ipc_router_lookup_remote_port(
server_port->server_addr.node_id,
server_port->server_addr.port_id);
if (!rport_ptr)
continue;
rport_ptr->sec_rule = rule;
}
up_read(&routing_table_lock_lha3);
server->synced_sec_rule = 1;
}
/**
* msm_ipc_sync_sec_rule() - Sync the security rule to the service
* @service: Service for which the rule has to be synchronized.
* @instance: Instance for which the rule has to be synchronized.
* @rule: Security rule to be synchronized.
*
* This function is used to syncrhonize the security rule with the server
* hash table, if the user-space script configures the rule after the service
* has come up. This function is used to synchronize the security rule to a
* specific service and optionally a specific instance.
*/
void msm_ipc_sync_sec_rule(uint32_t service, uint32_t instance, void *rule)
{
int key = (service & (SRV_HASH_SIZE - 1));
struct msm_ipc_server *server;
down_write(&server_list_lock_lha2);
list_for_each_entry(server, &server_list[key], list) {
if (server->name.service != service)
continue;
if (server->name.instance != instance &&
instance != ALL_INSTANCE)
continue;
/*
* If the rule applies to all instances and if the specific
* instance of a service has a rule synchronized already,
* do not apply the rule for that specific instance.
*/
if (instance == ALL_INSTANCE && server->synced_sec_rule)
continue;
sync_sec_rule(server, rule);
}
up_write(&server_list_lock_lha2);
}
/**
* msm_ipc_sync_default_sec_rule() - Default security rule to all services
* @rule: Security rule to be synchronized.
*
* This function is used to syncrhonize the security rule with the server
* hash table, if the user-space script configures the rule after the service
* has come up. This function is used to synchronize the security rule that
* applies to all services, if the concerned service do not have any rule
* defined.
*/
void msm_ipc_sync_default_sec_rule(void *rule)
{
int key;
struct msm_ipc_server *server;
down_write(&server_list_lock_lha2);
for (key = 0; key < SRV_HASH_SIZE; key++) {
list_for_each_entry(server, &server_list[key], list) {
if (server->synced_sec_rule)
continue;
sync_sec_rule(server, rule);
}
}
up_write(&server_list_lock_lha2);
}
static int process_hello_msg(struct msm_ipc_router_xprt_info *xprt_info,
struct rr_header_v1 *hdr)
{
int i, rc = 0;
union rr_control_msg ctl;
struct msm_ipc_routing_table_entry *rt_entry;
if (!hdr)
return -EINVAL;
RR("o HELLO NID %d\n", hdr->src_node_id);
xprt_info->remote_node_id = hdr->src_node_id;
/*
* Find the entry from Routing Table corresponding to Node ID.
* Under SSR, an entry will be found. When the system boots up
* for the 1st time, an entry will not be found and hence allocate
* an entry. Update the entry with the Node ID that it corresponds
* to and the XPRT through which it can be reached.
*/
down_write(&routing_table_lock_lha3);
rt_entry = lookup_routing_table(hdr->src_node_id);
if (!rt_entry) {
rt_entry = alloc_routing_table_entry(hdr->src_node_id);
if (!rt_entry) {
up_write(&routing_table_lock_lha3);
pr_err("%s: rt_entry allocation failed\n", __func__);
return -ENOMEM;
}
add_routing_table_entry(rt_entry);
}
down_write(&rt_entry->lock_lha4);
rt_entry->neighbor_node_id = xprt_info->remote_node_id;
rt_entry->xprt_info = xprt_info;
up_write(&rt_entry->lock_lha4);
up_write(&routing_table_lock_lha3);
/* Send a reply HELLO message */
memset(&ctl, 0, sizeof(ctl));
ctl.hello.cmd = IPC_ROUTER_CTRL_CMD_HELLO;
rc = msm_ipc_router_send_control_msg(xprt_info, &ctl,
IPC_ROUTER_DUMMY_DEST_NODE);
if (rc < 0) {
pr_err("%s: Error sending reply HELLO message\n", __func__);
return rc;
}
xprt_info->initialized = 1;
/*
* Send list of servers from the local node and from nodes
* outside the mesh network in which this XPRT is part of.
*/
down_read(&server_list_lock_lha2);
down_read(&routing_table_lock_lha3);
for (i = 0; i < RT_HASH_SIZE; i++) {
list_for_each_entry(rt_entry, &routing_table[i], list) {
if ((rt_entry->node_id != IPC_ROUTER_NID_LOCAL) &&
(!rt_entry->xprt_info ||
(rt_entry->xprt_info->xprt->link_id ==
xprt_info->xprt->link_id)))
continue;
rc = msm_ipc_router_send_server_list(rt_entry->node_id,
xprt_info);
if (rc < 0) {
up_read(&routing_table_lock_lha3);
up_read(&server_list_lock_lha2);
return rc;
}
}
}
up_read(&routing_table_lock_lha3);
up_read(&server_list_lock_lha2);
RR("HELLO message processed\n");
return rc;
}
static int process_resume_tx_msg(union rr_control_msg *msg,
struct rr_packet *pkt)
{
struct msm_ipc_router_remote_port *rport_ptr;
int ret = 0;
RR("o RESUME_TX id=%d:%08x\n", msg->cli.node_id, msg->cli.port_id);
down_read(&local_ports_lock_lha2);
down_read(&routing_table_lock_lha3);
rport_ptr = msm_ipc_router_lookup_remote_port(msg->cli.node_id,
msg->cli.port_id);
if (!rport_ptr) {
pr_err("%s: Unable to resume client\n", __func__);
ret = -ENODEV;
goto prtm_out;
}
mutex_lock(&rport_ptr->quota_lock_lhb2);
rport_ptr->tx_quota_cnt = 0;
post_resume_tx(rport_ptr, pkt);
mutex_unlock(&rport_ptr->quota_lock_lhb2);
prtm_out:
up_read(&routing_table_lock_lha3);
up_read(&local_ports_lock_lha2);
return 0;
}
static int process_new_server_msg(struct msm_ipc_router_xprt_info *xprt_info,
union rr_control_msg *msg, struct rr_packet *pkt)
{
struct msm_ipc_routing_table_entry *rt_entry;
struct msm_ipc_server *server;
struct msm_ipc_router_remote_port *rport_ptr;
if (msg->srv.instance == 0) {
pr_err("%s: Server %08x create rejected, version = 0\n",
__func__, msg->srv.service);
return -EINVAL;
}
RR("o NEW_SERVER id=%d:%08x service=%08x:%08x\n", msg->srv.node_id,
msg->srv.port_id, msg->srv.service, msg->srv.instance);
/*
* Find the entry from Routing Table corresponding to Node ID.
* Under SSR, an entry will be found. When the subsystem hosting
* service is not adjacent, an entry will not be found and hence
* allocate an entry. Update the entry with the Node ID that it
* corresponds to and the XPRT through which it can be reached.
*/
down_write(&routing_table_lock_lha3);
rt_entry = lookup_routing_table(msg->srv.node_id);
if (!rt_entry) {
rt_entry = alloc_routing_table_entry(msg->srv.node_id);
if (!rt_entry) {
up_write(&routing_table_lock_lha3);
pr_err("%s: rt_entry allocation failed\n", __func__);
return -ENOMEM;
}
down_write(&rt_entry->lock_lha4);
rt_entry->neighbor_node_id = xprt_info->remote_node_id;
rt_entry->xprt_info = xprt_info;
up_write(&rt_entry->lock_lha4);
add_routing_table_entry(rt_entry);
}
up_write(&routing_table_lock_lha3);
/*
* If the service does not exist already in the database, create and
* store the service info. Create a remote port structure in which
* the service is hosted and cache the security rule for the service
* in that remote port structure.
*/
down_write(&server_list_lock_lha2);
server = msm_ipc_router_lookup_server(msg->srv.service,
msg->srv.instance, msg->srv.node_id, msg->srv.port_id);
if (!server) {
server = msm_ipc_router_create_server(
msg->srv.service, msg->srv.instance,
msg->srv.node_id, msg->srv.port_id, xprt_info);
if (!server) {
up_write(&server_list_lock_lha2);
pr_err("%s: Server Create failed\n", __func__);
return -ENOMEM;
}
down_read(&routing_table_lock_lha3);
if (!msm_ipc_router_lookup_remote_port(
msg->srv.node_id, msg->srv.port_id)) {
rport_ptr = msm_ipc_router_create_remote_port(
msg->srv.node_id, msg->srv.port_id);
if (!rport_ptr) {
up_read(&routing_table_lock_lha3);
up_write(&server_list_lock_lha2);
return -ENOMEM;
}
rport_ptr->server = server;
rport_ptr->sec_rule = msm_ipc_get_security_rule(
msg->srv.service,
msg->srv.instance);
}
up_read(&routing_table_lock_lha3);
}
up_write(&server_list_lock_lha2);
/*
* Relay the new server message to other subsystems that do not belong
* to the cluster from which this message is received. Notify the
* local clients waiting for this service.
*/
relay_ctl_msg(xprt_info, msg);
post_control_ports(pkt);
return 0;
}
static int process_rmv_server_msg(struct msm_ipc_router_xprt_info *xprt_info,
union rr_control_msg *msg, struct rr_packet *pkt)
{
struct msm_ipc_server *server;
RR("o REMOVE_SERVER service=%08x:%d\n",
msg->srv.service, msg->srv.instance);
down_write(&server_list_lock_lha2);
server = msm_ipc_router_lookup_server(msg->srv.service,
msg->srv.instance, msg->srv.node_id, msg->srv.port_id);
if (server) {
msm_ipc_router_destroy_server(server, msg->srv.node_id,
msg->srv.port_id);
/*
* Relay the new server message to other subsystems that do not
* belong to the cluster from which this message is received.
* Notify the local clients communicating with the service.
*/
relay_ctl_msg(xprt_info, msg);
post_control_ports(pkt);
}
up_write(&server_list_lock_lha2);
return 0;
}
static int process_rmv_client_msg(struct msm_ipc_router_xprt_info *xprt_info,
union rr_control_msg *msg, struct rr_packet *pkt)
{
struct msm_ipc_router_remote_port *rport_ptr;
RR("o REMOVE_CLIENT id=%d:%08x\n", msg->cli.node_id, msg->cli.port_id);
down_write(&routing_table_lock_lha3);
rport_ptr = msm_ipc_router_lookup_remote_port(msg->cli.node_id,
msg->cli.port_id);
if (rport_ptr)
msm_ipc_router_destroy_remote_port(rport_ptr);
up_write(&routing_table_lock_lha3);
relay_ctl_msg(xprt_info, msg);
post_control_ports(pkt);
return 0;
}
static int process_control_msg(struct msm_ipc_router_xprt_info *xprt_info,
struct rr_packet *pkt)
{
union rr_control_msg *msg;
int rc = 0;
struct rr_header_v1 *hdr;
if (pkt->length != sizeof(*msg)) {
pr_err("%s: r2r msg size %d != %d\n", __func__, pkt->length,
sizeof(*msg));
return -EINVAL;
}
hdr = &(pkt->hdr);
msg = msm_ipc_router_skb_to_buf(pkt->pkt_fragment_q, sizeof(*msg));
if (!msg) {
pr_err("%s: Error extracting control msg\n", __func__);
return -ENOMEM;
}
switch (msg->cmd) {
case IPC_ROUTER_CTRL_CMD_HELLO:
rc = process_hello_msg(xprt_info, hdr);
break;
case IPC_ROUTER_CTRL_CMD_RESUME_TX:
rc = process_resume_tx_msg(msg, pkt);
break;
case IPC_ROUTER_CTRL_CMD_NEW_SERVER:
rc = process_new_server_msg(xprt_info, msg, pkt);
break;
case IPC_ROUTER_CTRL_CMD_REMOVE_SERVER:
rc = process_rmv_server_msg(xprt_info, msg, pkt);
break;
case IPC_ROUTER_CTRL_CMD_REMOVE_CLIENT:
rc = process_rmv_client_msg(xprt_info, msg, pkt);
break;
case IPC_ROUTER_CTRL_CMD_PING:
/* No action needed for ping messages received */
RR("o PING\n");
break;
default:
RR("o UNKNOWN(%08x)\n", msg->cmd);
rc = -ENOSYS;
}
kfree(msg);
return rc;
}
static void do_read_data(struct work_struct *work)
{
struct rr_header_v1 *hdr;
struct rr_packet *pkt = NULL;
struct msm_ipc_port *port_ptr;
struct msm_ipc_router_remote_port *rport_ptr;
int ret;
struct msm_ipc_router_xprt_info *xprt_info =
container_of(work,
struct msm_ipc_router_xprt_info,
read_data);
while ((pkt = rr_read(xprt_info)) != NULL) {
if (pkt->length < IPC_ROUTER_HDR_SIZE ||
pkt->length > MAX_IPC_PKT_SIZE) {
pr_err("%s: Invalid pkt length %d\n",
__func__, pkt->length);
goto fail_data;
}
ret = extract_header(pkt);
if (ret < 0)
goto fail_data;
hdr = &(pkt->hdr);
RAW("ver=%d type=%d src=%d:%08x crx=%d siz=%d dst=%d:%08x\n",
hdr->version, hdr->type, hdr->src_node_id,
hdr->src_port_id, hdr->control_flag, hdr->size,
hdr->dst_node_id, hdr->dst_port_id);
if ((hdr->dst_node_id != IPC_ROUTER_NID_LOCAL) &&
((hdr->type == IPC_ROUTER_CTRL_CMD_RESUME_TX) ||
(hdr->type == IPC_ROUTER_CTRL_CMD_DATA))) {
forward_msg(xprt_info, pkt);
release_pkt(pkt);
continue;
}
if (hdr->type != IPC_ROUTER_CTRL_CMD_DATA) {
process_control_msg(xprt_info, pkt);
release_pkt(pkt);
continue;
}
#if defined(CONFIG_MSM_SMD_LOGGING)
#if defined(DEBUG)
if (msm_ipc_router_debug_mask & SMEM_LOG) {
smem_log_event((SMEM_LOG_PROC_ID_APPS |
SMEM_LOG_IPC_ROUTER_EVENT_BASE |
IPC_ROUTER_LOG_EVENT_RX),
(hdr->src_node_id << 24) |
(hdr->src_port_id & 0xffffff),
(hdr->dst_node_id << 24) |
(hdr->dst_port_id & 0xffffff),
(hdr->type << 24) | (hdr->control_flag << 16) |
(hdr->size & 0xffff));
}
#endif
#endif
down_read(&local_ports_lock_lha2);
port_ptr = msm_ipc_router_lookup_local_port(hdr->dst_port_id);
if (!port_ptr) {
pr_err("%s: No local port id %08x\n", __func__,
hdr->dst_port_id);
up_read(&local_ports_lock_lha2);
release_pkt(pkt);
return;
}
down_read(&routing_table_lock_lha3);
rport_ptr = msm_ipc_router_lookup_remote_port(hdr->src_node_id,
hdr->src_port_id);
if (!rport_ptr) {
rport_ptr = msm_ipc_router_create_remote_port(
hdr->src_node_id,
hdr->src_port_id);
if (!rport_ptr) {
pr_err("%s: Rmt Prt %08x:%08x create failed\n",
__func__, hdr->src_node_id,
hdr->src_port_id);
up_read(&routing_table_lock_lha3);
up_read(&local_ports_lock_lha2);
release_pkt(pkt);
return;
}
}
up_read(&routing_table_lock_lha3);
post_pkt_to_port(port_ptr, pkt, 0);
up_read(&local_ports_lock_lha2);
}
return;
fail_data:
release_pkt(pkt);
pr_err("ipc_router has died\n");
}
int msm_ipc_router_register_server(struct msm_ipc_port *port_ptr,
struct msm_ipc_addr *name)
{
struct msm_ipc_server *server;
unsigned long flags;
union rr_control_msg ctl;
if (!port_ptr || !name)
return -EINVAL;
if (name->addrtype != MSM_IPC_ADDR_NAME)
return -EINVAL;
down_write(&server_list_lock_lha2);
server = msm_ipc_router_lookup_server(name->addr.port_name.service,
name->addr.port_name.instance,
IPC_ROUTER_NID_LOCAL,
port_ptr->this_port.port_id);
if (server) {
up_write(&server_list_lock_lha2);
pr_err("%s: Server already present\n", __func__);
return -EINVAL;
}
server = msm_ipc_router_create_server(name->addr.port_name.service,
name->addr.port_name.instance,
IPC_ROUTER_NID_LOCAL,
port_ptr->this_port.port_id,
NULL);
if (!server) {
up_write(&server_list_lock_lha2);
pr_err("%s: Server Creation failed\n", __func__);
return -EINVAL;
}
memset(&ctl, 0, sizeof(ctl));
ctl.cmd = IPC_ROUTER_CTRL_CMD_NEW_SERVER;
ctl.srv.service = server->name.service;
ctl.srv.instance = server->name.instance;
ctl.srv.node_id = IPC_ROUTER_NID_LOCAL;
ctl.srv.port_id = port_ptr->this_port.port_id;
up_write(&server_list_lock_lha2);
broadcast_ctl_msg(&ctl);
broadcast_ctl_msg_locally(&ctl);
spin_lock_irqsave(&port_ptr->port_lock, flags);
port_ptr->type = SERVER_PORT;
port_ptr->mode_info.mode = MULTI_LINK_MODE;
port_ptr->port_name.service = server->name.service;
port_ptr->port_name.instance = server->name.instance;
spin_unlock_irqrestore(&port_ptr->port_lock, flags);
return 0;
}
int msm_ipc_router_unregister_server(struct msm_ipc_port *port_ptr)
{
struct msm_ipc_server *server;
unsigned long flags;
union rr_control_msg ctl;
if (!port_ptr)
return -EINVAL;
if (port_ptr->type != SERVER_PORT) {
pr_err("%s: Trying to unregister a non-server port\n",
__func__);
return -EINVAL;
}
if (port_ptr->this_port.node_id != IPC_ROUTER_NID_LOCAL) {
pr_err("%s: Trying to unregister a remote server locally\n",
__func__);
return -EINVAL;
}
down_write(&server_list_lock_lha2);
server = msm_ipc_router_lookup_server(port_ptr->port_name.service,
port_ptr->port_name.instance,
port_ptr->this_port.node_id,
port_ptr->this_port.port_id);
if (!server) {
up_write(&server_list_lock_lha2);
pr_err("%s: Server lookup failed\n", __func__);
return -ENODEV;
}
memset(&ctl, 0, sizeof(ctl));
ctl.cmd = IPC_ROUTER_CTRL_CMD_REMOVE_SERVER;
ctl.srv.service = server->name.service;
ctl.srv.instance = server->name.instance;
ctl.srv.node_id = IPC_ROUTER_NID_LOCAL;
ctl.srv.port_id = port_ptr->this_port.port_id;
msm_ipc_router_destroy_server(server, port_ptr->this_port.node_id,
port_ptr->this_port.port_id);
up_write(&server_list_lock_lha2);
broadcast_ctl_msg(&ctl);
broadcast_ctl_msg_locally(&ctl);
spin_lock_irqsave(&port_ptr->port_lock, flags);
port_ptr->type = CLIENT_PORT;
spin_unlock_irqrestore(&port_ptr->port_lock, flags);
return 0;
}
static int loopback_data(struct msm_ipc_port *src,
uint32_t port_id,
struct sk_buff_head *data)
{
struct rr_header_v1 *hdr;
struct msm_ipc_port *port_ptr;
struct rr_packet *pkt;
int ret_len;
if (!data) {
pr_err("%s: Invalid pkt pointer\n", __func__);
return -EINVAL;
}
pkt = create_pkt(data);
if (!pkt) {
pr_err("%s: New pkt create failed\n", __func__);
return -ENOMEM;
}
hdr = &(pkt->hdr);
hdr->version = IPC_ROUTER_V1;
hdr->type = IPC_ROUTER_CTRL_CMD_DATA;
hdr->src_node_id = src->this_port.node_id;
hdr->src_port_id = src->this_port.port_id;
hdr->size = pkt->length;
hdr->control_flag = 0;
hdr->dst_node_id = IPC_ROUTER_NID_LOCAL;
hdr->dst_port_id = port_id;
down_read(&local_ports_lock_lha2);
port_ptr = msm_ipc_router_lookup_local_port(port_id);
if (!port_ptr) {
pr_err("%s: Local port %d not present\n", __func__, port_id);
up_read(&local_ports_lock_lha2);
pkt->pkt_fragment_q = NULL;
release_pkt(pkt);
return -ENODEV;
}
ret_len = pkt->length;
post_pkt_to_port(port_ptr, pkt, 0);
update_comm_mode_info(&src->mode_info, NULL);
up_read(&local_ports_lock_lha2);
return ret_len;
}
static int msm_ipc_router_write_pkt(struct msm_ipc_port *src,
struct msm_ipc_router_remote_port *rport_ptr,
struct rr_packet *pkt)
{
struct rr_header_v1 *hdr;
struct msm_ipc_router_xprt_info *xprt_info;
struct msm_ipc_routing_table_entry *rt_entry;
struct msm_ipc_resume_tx_port *resume_tx_port;
struct sk_buff *temp_skb;
int xprt_option;
int ret;
int align_size;
if (!rport_ptr || !src || !pkt)
return -EINVAL;
hdr = &(pkt->hdr);
hdr->type = IPC_ROUTER_CTRL_CMD_DATA;
hdr->src_node_id = src->this_port.node_id;
hdr->src_port_id = src->this_port.port_id;
hdr->size = pkt->length;
hdr->control_flag = 0;
hdr->dst_node_id = rport_ptr->node_id;
hdr->dst_port_id = rport_ptr->port_id;
mutex_lock(&rport_ptr->quota_lock_lhb2);
if (rport_ptr->tx_quota_cnt == IPC_ROUTER_DEFAULT_RX_QUOTA) {
if (msm_ipc_router_lookup_resume_tx_port(
rport_ptr, src->this_port.port_id)) {
mutex_unlock(&rport_ptr->quota_lock_lhb2);
return -EAGAIN;
}
resume_tx_port =
kzalloc(sizeof(struct msm_ipc_resume_tx_port),
GFP_KERNEL);
if (!resume_tx_port) {
pr_err("%s: Resume_Tx port allocation failed\n",
__func__);
mutex_unlock(&rport_ptr->quota_lock_lhb2);
return -ENOMEM;
}
INIT_LIST_HEAD(&resume_tx_port->list);
resume_tx_port->port_id = src->this_port.port_id;
resume_tx_port->node_id = src->this_port.node_id;
list_add_tail(&resume_tx_port->list,
&rport_ptr->resume_tx_port_list);
mutex_unlock(&rport_ptr->quota_lock_lhb2);
return -EAGAIN;
}
rport_ptr->tx_quota_cnt++;
if (rport_ptr->tx_quota_cnt == IPC_ROUTER_DEFAULT_RX_QUOTA)
hdr->control_flag |= CONTROL_FLAG_CONFIRM_RX;
mutex_unlock(&rport_ptr->quota_lock_lhb2);
rt_entry = lookup_routing_table(hdr->dst_node_id);
if (!rt_entry || !rt_entry->xprt_info) {
pr_err("%s: Remote node %d not up\n",
__func__, hdr->dst_node_id);
return -ENODEV;
}
down_read(&rt_entry->lock_lha4);
xprt_info = rt_entry->xprt_info;
ret = prepend_header(pkt, xprt_info);
if (ret < 0) {
up_read(&rt_entry->lock_lha4);
pr_err("%s: Prepend Header failed\n", __func__);
return ret;
}
xprt_option = xprt_info->xprt->get_option(xprt_info->xprt);
if (!(xprt_option & FRAG_PKT_WRITE_ENABLE)) {
ret = defragment_pkt(pkt);
if (ret < 0) {
up_read(&rt_entry->lock_lha4);
return ret;
}
}
temp_skb = skb_peek_tail(pkt->pkt_fragment_q);
align_size = ALIGN_SIZE(pkt->length);
skb_put(temp_skb, align_size);
pkt->length += align_size;
mutex_lock(&xprt_info->tx_lock_lhb2);
ret = xprt_info->xprt->write(pkt, pkt->length, xprt_info->xprt);
mutex_unlock(&xprt_info->tx_lock_lhb2);
up_read(&rt_entry->lock_lha4);
if (ret < 0) {
pr_err("%s: Write on XPRT failed\n", __func__);
return ret;
}
update_comm_mode_info(&src->mode_info, xprt_info);
RAW_HDR("[w rr_h] "
"ver=%i,type=%s,src_nid=%08x,src_port_id=%08x,"
"control_flag=%i,size=%3i,dst_pid=%08x,dst_cid=%08x\n",
hdr->version, type_to_str(hdr->type),
hdr->src_node_id, hdr->src_port_id,
hdr->control_flag, hdr->size,
hdr->dst_node_id, hdr->dst_port_id);
#if defined(CONFIG_MSM_SMD_LOGGING)
#if defined(DEBUG)
if (msm_ipc_router_debug_mask & SMEM_LOG) {
smem_log_event((SMEM_LOG_PROC_ID_APPS |
SMEM_LOG_IPC_ROUTER_EVENT_BASE |
IPC_ROUTER_LOG_EVENT_TX),
(hdr->src_node_id << 24) |
(hdr->src_port_id & 0xffffff),
(hdr->dst_node_id << 24) |
(hdr->dst_port_id & 0xffffff),
(hdr->type << 24) | (hdr->control_flag << 16) |
(hdr->size & 0xffff));
}
#endif
#endif
return hdr->size;
}
int msm_ipc_router_send_to(struct msm_ipc_port *src,
struct sk_buff_head *data,
struct msm_ipc_addr *dest)
{
uint32_t dst_node_id = 0, dst_port_id = 0;
struct msm_ipc_server *server;
struct msm_ipc_server_port *server_port;
struct msm_ipc_router_remote_port *rport_ptr = NULL;
struct rr_packet *pkt;
int ret;
if (!src || !data || !dest) {
pr_err("%s: Invalid Parameters\n", __func__);
return -EINVAL;
}
/* Resolve Address*/
if (dest->addrtype == MSM_IPC_ADDR_ID) {
dst_node_id = dest->addr.port_addr.node_id;
dst_port_id = dest->addr.port_addr.port_id;
} else if (dest->addrtype == MSM_IPC_ADDR_NAME) {
down_read(&server_list_lock_lha2);
server = msm_ipc_router_lookup_server(
dest->addr.port_name.service,
dest->addr.port_name.instance,
0, 0);
if (!server) {
up_read(&server_list_lock_lha2);
pr_err("%s: Destination not reachable\n", __func__);
return -ENODEV;
}
server_port = list_first_entry(&server->server_port_list,
struct msm_ipc_server_port,
list);
dst_node_id = server_port->server_addr.node_id;
dst_port_id = server_port->server_addr.port_id;
up_read(&server_list_lock_lha2);
}
if (dst_node_id == IPC_ROUTER_NID_LOCAL) {
ret = loopback_data(src, dst_port_id, data);
return ret;
}
down_read(&routing_table_lock_lha3);
rport_ptr = msm_ipc_router_lookup_remote_port(dst_node_id,
dst_port_id);
if (!rport_ptr) {
up_read(&routing_table_lock_lha3);
pr_err("%s: Remote port not found\n", __func__);
return -ENODEV;
}
if (src->check_send_permissions) {
ret = src->check_send_permissions(rport_ptr->sec_rule);
if (ret <= 0) {
up_read(&routing_table_lock_lha3);
pr_err("%s: permission failure for %s\n",
__func__, current->comm);
return -EPERM;
}
}
pkt = create_pkt(data);
if (!pkt) {
up_read(&routing_table_lock_lha3);
pr_err("%s: Pkt creation failed\n", __func__);
return -ENOMEM;
}
ret = msm_ipc_router_write_pkt(src, rport_ptr, pkt);
up_read(&routing_table_lock_lha3);
if (ret < 0)
pkt->pkt_fragment_q = NULL;
release_pkt(pkt);
return ret;
}
int msm_ipc_router_send_msg(struct msm_ipc_port *src,
struct msm_ipc_addr *dest,
void *data, unsigned int data_len)
{
struct sk_buff_head *out_skb_head;
int ret;
out_skb_head = msm_ipc_router_buf_to_skb(data, data_len);
if (!out_skb_head) {
pr_err("%s: SKB conversion failed\n", __func__);
return -EFAULT;
}
ret = msm_ipc_router_send_to(src, out_skb_head, dest);
if (ret < 0) {
if (ret != -EAGAIN)
pr_err("%s: msm_ipc_router_send_to failed - ret: %d\n",
__func__, ret);
msm_ipc_router_free_skb(out_skb_head);
return ret;
}
return 0;
}
/**
* msm_ipc_router_send_resume_tx() - Send Resume_Tx message
* @data: Pointer to received data packet that has confirm_rx bit set
*
* @return: On success, number of bytes transferred is returned, else
* standard linux error code is returned.
*
* This function sends the Resume_Tx event to the remote node that
* sent the data with confirm_rx field set. In case of a multi-hop
* scenario also, this function makes sure that the destination node_id
* to which the resume_tx event should reach is right.
*/
static int msm_ipc_router_send_resume_tx(void *data)
{
union rr_control_msg msg;
struct rr_header_v1 *hdr = (struct rr_header_v1 *)data;
struct msm_ipc_routing_table_entry *rt_entry;
int ret;
memset(&msg, 0, sizeof(msg));
msg.cmd = IPC_ROUTER_CTRL_CMD_RESUME_TX;
msg.cli.node_id = hdr->dst_node_id;
msg.cli.port_id = hdr->dst_port_id;
down_read(&routing_table_lock_lha3);
rt_entry = lookup_routing_table(hdr->src_node_id);
if (!rt_entry) {
pr_err("%s: %d Node is not present",
__func__, hdr->src_node_id);
up_read(&routing_table_lock_lha3);
return -ENODEV;
}
RR("x RESUME_TX id=%d:%08x\n",
msg.cli.node_id, msg.cli.port_id);
ret = msm_ipc_router_send_control_msg(rt_entry->xprt_info, &msg,
hdr->src_node_id);
up_read(&routing_table_lock_lha3);
if (ret < 0)
pr_err("%s: Send Resume_Tx Failed SRC_NODE: %d SRC_PORT: %d DEST_NODE: %d",
__func__, hdr->dst_node_id, hdr->dst_port_id,
hdr->src_node_id);
return ret;
}
int msm_ipc_router_read(struct msm_ipc_port *port_ptr,
struct rr_packet **read_pkt,
size_t buf_len)
{
struct rr_packet *pkt;
if (!port_ptr || !read_pkt)
return -EINVAL;
mutex_lock(&port_ptr->port_rx_q_lock_lhb3);
if (list_empty(&port_ptr->port_rx_q)) {
mutex_unlock(&port_ptr->port_rx_q_lock_lhb3);
return -EAGAIN;
}
pkt = list_first_entry(&port_ptr->port_rx_q, struct rr_packet, list);
if ((buf_len) && (pkt->hdr.size > buf_len)) {
mutex_unlock(&port_ptr->port_rx_q_lock_lhb3);
return -ETOOSMALL;
}
list_del(&pkt->list);
if (list_empty(&port_ptr->port_rx_q))
wake_unlock(&port_ptr->port_rx_wake_lock);
*read_pkt = pkt;
mutex_unlock(&port_ptr->port_rx_q_lock_lhb3);
if (pkt->hdr.control_flag & CONTROL_FLAG_CONFIRM_RX)
msm_ipc_router_send_resume_tx(&pkt->hdr);
return pkt->length;
}
/**
* msm_ipc_router_rx_data_wait() - Wait for new message destined to a local port.
* @port_ptr: Pointer to the local port
* @timeout: < 0 timeout indicates infinite wait till a message arrives.
* > 0 timeout indicates the wait time.
* 0 indicates that we do not wait.
* @return: 0 if there are pending messages to read,
* standard Linux error code otherwise.
*
* Checks for the availability of messages that are destined to a local port.
* If no messages are present then waits as per @timeout.
*/
int msm_ipc_router_rx_data_wait(struct msm_ipc_port *port_ptr, long timeout)
{
int ret = 0;
mutex_lock(&port_ptr->port_rx_q_lock_lhb3);
while (list_empty(&port_ptr->port_rx_q)) {
mutex_unlock(&port_ptr->port_rx_q_lock_lhb3);
if (timeout < 0) {
ret = wait_event_interruptible(
port_ptr->port_rx_wait_q,
!list_empty(&port_ptr->port_rx_q));
if (ret)
return ret;
} else if (timeout > 0) {
timeout = wait_event_interruptible_timeout(
port_ptr->port_rx_wait_q,
!list_empty(&port_ptr->port_rx_q),
timeout);
if (timeout < 0)
return -EFAULT;
}
if (timeout == 0)
return -ENOMSG;
mutex_lock(&port_ptr->port_rx_q_lock_lhb3);
}
mutex_unlock(&port_ptr->port_rx_q_lock_lhb3);
return ret;
}
/**
* msm_ipc_router_recv_from() - Recieve messages destined to a local port.
* @port_ptr: Pointer to the local port
* @pkt : Pointer to the router-to-router packet
* @src: Pointer to local port address
* @timeout: < 0 timeout indicates infinite wait till a message arrives.
* > 0 timeout indicates the wait time.
* 0 indicates that we do not wait.
* @return: = Number of bytes read(On successful read operation).
* = -ENOMSG (If there are no pending messages and timeout is 0).
* = -EINVAL (If either of the arguments, port_ptr or data is invalid)
* = -EFAULT (If there are no pending messages when timeout is > 0
* and the wait_event_interruptible_timeout has returned value > 0)
* = -ERESTARTSYS (If there are no pending messages when timeout
* is < 0 and wait_event_interruptible was interrupted by a signal)
*
* This function reads the messages that are destined for a local port. It
* is used by modules that exist with-in the kernel and use IPC Router for
* transport. The function checks if there are any messages that are already
* received. If yes, it reads them, else it waits as per the timeout value.
* On a successful read, the return value of the function indicates the number
* of bytes that are read.
*/
int msm_ipc_router_recv_from(struct msm_ipc_port *port_ptr,
struct rr_packet **pkt,
struct msm_ipc_addr *src,
long timeout)
{
int ret, data_len, align_size;
struct sk_buff *temp_skb;
struct rr_header_v1 *hdr = NULL;
if (!port_ptr || !pkt) {
pr_err("%s: Invalid pointers being passed\n", __func__);
return -EINVAL;
}
*pkt = NULL;
ret = msm_ipc_router_rx_data_wait(port_ptr, timeout);
if (ret)
return ret;
ret = msm_ipc_router_read(port_ptr, pkt, 0);
if (ret <= 0 || !(*pkt))
return ret;
hdr = &((*pkt)->hdr);
if (src) {
src->addrtype = MSM_IPC_ADDR_ID;
src->addr.port_addr.node_id = hdr->src_node_id;
src->addr.port_addr.port_id = hdr->src_port_id;
}
data_len = hdr->size;
align_size = ALIGN_SIZE(data_len);
if (align_size) {
temp_skb = skb_peek_tail((*pkt)->pkt_fragment_q);
skb_trim(temp_skb, (temp_skb->len - align_size));
}
return data_len;
}
int msm_ipc_router_read_msg(struct msm_ipc_port *port_ptr,
struct msm_ipc_addr *src,
unsigned char **data,
unsigned int *len)
{
struct rr_packet *pkt;
int ret;
ret = msm_ipc_router_recv_from(port_ptr, &pkt, src, 0);
if (ret < 0) {
if (ret != -ENOMSG)
pr_err("%s: msm_ipc_router_recv_from failed - ret: %d\n",
__func__, ret);
return ret;
}
*data = msm_ipc_router_skb_to_buf(pkt->pkt_fragment_q, ret);
if (!(*data))
pr_err("%s: Buf conversion failed\n", __func__);
*len = ret;
release_pkt(pkt);
return 0;
}
struct msm_ipc_port *msm_ipc_router_create_port(
void (*notify)(unsigned event, void *priv),
void *priv)
{
struct msm_ipc_port *port_ptr;
int ret;
ret = wait_for_completion_interruptible(&msm_ipc_local_router_up);
if (ret < 0) {
pr_err("%s: Error waiting for local router\n", __func__);
return NULL;
}
port_ptr = msm_ipc_router_create_raw_port(NULL, notify, priv);
if (!port_ptr)
pr_err("%s: port_ptr alloc failed\n", __func__);
return port_ptr;
}
int msm_ipc_router_close_port(struct msm_ipc_port *port_ptr)
{
union rr_control_msg msg;
struct rr_packet *pkt, *temp_pkt;
struct msm_ipc_server *server;
if (!port_ptr)
return -EINVAL;
if (port_ptr->type == SERVER_PORT || port_ptr->type == CLIENT_PORT) {
down_write(&local_ports_lock_lha2);
list_del(&port_ptr->list);
up_write(&local_ports_lock_lha2);
if (port_ptr->type == SERVER_PORT) {
memset(&msg, 0, sizeof(msg));
msg.cmd = IPC_ROUTER_CTRL_CMD_REMOVE_SERVER;
msg.srv.service = port_ptr->port_name.service;
msg.srv.instance = port_ptr->port_name.instance;
msg.srv.node_id = port_ptr->this_port.node_id;
msg.srv.port_id = port_ptr->this_port.port_id;
RR("x REMOVE_SERVER Name=%d:%08x Id=%d:%08x\n",
msg.srv.service, msg.srv.instance,
msg.srv.node_id, msg.srv.port_id);
broadcast_ctl_msg(&msg);
broadcast_ctl_msg_locally(&msg);
}
/*
* Server port could have been a client port earlier.
* Send REMOVE_CLIENT message in either case.
*/
RR("x REMOVE_CLIENT id=%d:%08x\n",
port_ptr->this_port.node_id, port_ptr->this_port.port_id);
msm_ipc_router_send_remove_client(&port_ptr->mode_info,
port_ptr->this_port.node_id,
port_ptr->this_port.port_id);
} else if (port_ptr->type == CONTROL_PORT) {
down_write(&control_ports_lock_lha5);
list_del(&port_ptr->list);
up_write(&control_ports_lock_lha5);
} else if (port_ptr->type == IRSC_PORT) {
down_write(&local_ports_lock_lha2);
list_del(&port_ptr->list);
up_write(&local_ports_lock_lha2);
signal_irsc_completion();
}
mutex_lock(&port_ptr->port_rx_q_lock_lhb3);
list_for_each_entry_safe(pkt, temp_pkt, &port_ptr->port_rx_q, list) {
list_del(&pkt->list);
release_pkt(pkt);
}
mutex_unlock(&port_ptr->port_rx_q_lock_lhb3);
if (port_ptr->type == SERVER_PORT) {
down_write(&server_list_lock_lha2);
server = msm_ipc_router_lookup_server(
port_ptr->port_name.service,
port_ptr->port_name.instance,
port_ptr->this_port.node_id,
port_ptr->this_port.port_id);
if (server)
msm_ipc_router_destroy_server(server,
port_ptr->this_port.node_id,
port_ptr->this_port.port_id);
up_write(&server_list_lock_lha2);
}
wake_lock_destroy(&port_ptr->port_rx_wake_lock);
kfree(port_ptr);
return 0;
}
int msm_ipc_router_get_curr_pkt_size(struct msm_ipc_port *port_ptr)
{
struct rr_packet *pkt;
int rc = 0;
if (!port_ptr)
return -EINVAL;
mutex_lock(&port_ptr->port_rx_q_lock_lhb3);
if (!list_empty(&port_ptr->port_rx_q)) {
pkt = list_first_entry(&port_ptr->port_rx_q,
struct rr_packet, list);
rc = pkt->length;
}
mutex_unlock(&port_ptr->port_rx_q_lock_lhb3);
return rc;
}
int msm_ipc_router_bind_control_port(struct msm_ipc_port *port_ptr)
{
if (!port_ptr)
return -EINVAL;
down_write(&local_ports_lock_lha2);
list_del(&port_ptr->list);
up_write(&local_ports_lock_lha2);
port_ptr->type = CONTROL_PORT;
down_write(&control_ports_lock_lha5);
list_add_tail(&port_ptr->list, &control_ports);
up_write(&control_ports_lock_lha5);
return 0;
}
int msm_ipc_router_lookup_server_name(struct msm_ipc_port_name *srv_name,
struct msm_ipc_server_info *srv_info,
int num_entries_in_array,
uint32_t lookup_mask)
{
struct msm_ipc_server *server;
struct msm_ipc_server_port *server_port;
int key, i = 0; /*num_entries_found*/
if (!srv_name) {
pr_err("%s: Invalid srv_name\n", __func__);
return -EINVAL;
}
if (num_entries_in_array && !srv_info) {
pr_err("%s: srv_info NULL\n", __func__);
return -EINVAL;
}
down_read(&server_list_lock_lha2);
if (!lookup_mask)
lookup_mask = 0xFFFFFFFF;
key = (srv_name->service & (SRV_HASH_SIZE - 1));
list_for_each_entry(server, &server_list[key], list) {
if ((server->name.service != srv_name->service) ||
((server->name.instance & lookup_mask) !=
srv_name->instance))
continue;
list_for_each_entry(server_port,
&server->server_port_list, list) {
if (i < num_entries_in_array) {
srv_info[i].node_id =
server_port->server_addr.node_id;
srv_info[i].port_id =
server_port->server_addr.port_id;
srv_info[i].service = server->name.service;
srv_info[i].instance = server->name.instance;
}
i++;
}
}
up_read(&server_list_lock_lha2);
return i;
}
int msm_ipc_router_close(void)
{
struct msm_ipc_router_xprt_info *xprt_info, *tmp_xprt_info;
down_write(&xprt_info_list_lock_lha5);
list_for_each_entry_safe(xprt_info, tmp_xprt_info,
&xprt_info_list, list) {
xprt_info->xprt->close(xprt_info->xprt);
list_del(&xprt_info->list);
kfree(xprt_info);
}
up_write(&xprt_info_list_lock_lha5);
return 0;
}
#if defined(CONFIG_DEBUG_FS)
static int dump_routing_table(char *buf, int max)
{
int i = 0, j;
struct msm_ipc_routing_table_entry *rt_entry;
for (j = 0; j < RT_HASH_SIZE; j++) {
down_read(&routing_table_lock_lha3);
list_for_each_entry(rt_entry, &routing_table[j], list) {
down_read(&rt_entry->lock_lha4);
i += scnprintf(buf + i, max - i,
"Node Id: 0x%08x\n", rt_entry->node_id);
if (rt_entry->node_id == IPC_ROUTER_NID_LOCAL) {
i += scnprintf(buf + i, max - i,
"XPRT Name: Loopback\n");
i += scnprintf(buf + i, max - i,
"Next Hop: %d\n", rt_entry->node_id);
} else {
i += scnprintf(buf + i, max - i,
"XPRT Name: %s\n",
rt_entry->xprt_info->xprt->name);
i += scnprintf(buf + i, max - i,
"Next Hop: 0x%08x\n",
rt_entry->xprt_info->remote_node_id);
}
i += scnprintf(buf + i, max - i, "\n");
up_read(&rt_entry->lock_lha4);
}
up_read(&routing_table_lock_lha3);
}
return i;
}
static int dump_xprt_info(char *buf, int max)
{
int i = 0;
struct msm_ipc_router_xprt_info *xprt_info;
down_read(&xprt_info_list_lock_lha5);
list_for_each_entry(xprt_info, &xprt_info_list, list) {
i += scnprintf(buf + i, max - i, "XPRT Name: %s\n",
xprt_info->xprt->name);
i += scnprintf(buf + i, max - i, "Link Id: %d\n",
xprt_info->xprt->link_id);
i += scnprintf(buf + i, max - i, "Initialized: %s\n",
(xprt_info->initialized ? "Y" : "N"));
i += scnprintf(buf + i, max - i, "Remote Node Id: 0x%08x\n",
xprt_info->remote_node_id);
i += scnprintf(buf + i, max - i, "\n");
}
up_read(&xprt_info_list_lock_lha5);
return i;
}
static int dump_servers(char *buf, int max)
{
int i = 0, j;
struct msm_ipc_server *server;
struct msm_ipc_server_port *server_port;
down_read(&server_list_lock_lha2);
for (j = 0; j < SRV_HASH_SIZE; j++) {
list_for_each_entry(server, &server_list[j], list) {
list_for_each_entry(server_port,
&server->server_port_list,
list) {
i += scnprintf(buf + i, max - i, "Service: "
"0x%08x\n", server->name.service);
i += scnprintf(buf + i, max - i, "Instance: "
"0x%08x\n", server->name.instance);
i += scnprintf(buf + i, max - i,
"Node_id: 0x%08x\n",
server_port->server_addr.node_id);
i += scnprintf(buf + i, max - i,
"Port_id: 0x%08x\n",
server_port->server_addr.port_id);
i += scnprintf(buf + i, max - i, "\n");
}
}
}
up_read(&server_list_lock_lha2);
return i;
}
static int dump_remote_ports(char *buf, int max)
{
int i = 0, j, k;
struct msm_ipc_router_remote_port *rport_ptr;
struct msm_ipc_routing_table_entry *rt_entry;
for (j = 0; j < RT_HASH_SIZE; j++) {
down_read(&routing_table_lock_lha3);
list_for_each_entry(rt_entry, &routing_table[j], list) {
down_read(&rt_entry->lock_lha4);
for (k = 0; k < RP_HASH_SIZE; k++) {
list_for_each_entry(rport_ptr,
&rt_entry->remote_port_list[k],
list) {
i += scnprintf(buf + i, max - i,
"Node_id: 0x%08x\n",
rport_ptr->node_id);
i += scnprintf(buf + i, max - i,
"Port_id: 0x%08x\n",
rport_ptr->port_id);
i += scnprintf(buf + i, max - i,
"Quota_cnt: %d\n",
rport_ptr->tx_quota_cnt);
i += scnprintf(buf + i, max - i, "\n");
}
}
up_read(&rt_entry->lock_lha4);
}
up_read(&routing_table_lock_lha3);
}
return i;
}
static int dump_control_ports(char *buf, int max)
{
int i = 0;
struct msm_ipc_port *port_ptr;
down_read(&control_ports_lock_lha5);
list_for_each_entry(port_ptr, &control_ports, list) {
i += scnprintf(buf + i, max - i, "Node_id: 0x%08x\n",
port_ptr->this_port.node_id);
i += scnprintf(buf + i, max - i, "Port_id: 0x%08x\n",
port_ptr->this_port.port_id);
i += scnprintf(buf + i, max - i, "\n");
}
up_read(&control_ports_lock_lha5);
return i;
}
static int dump_local_ports(char *buf, int max)
{
int i = 0, j;
unsigned long flags;
struct msm_ipc_port *port_ptr;
down_read(&local_ports_lock_lha2);
for (j = 0; j < LP_HASH_SIZE; j++) {
list_for_each_entry(port_ptr, &local_ports[j], list) {
spin_lock_irqsave(&port_ptr->port_lock, flags);
i += scnprintf(buf + i, max - i, "Node_id: 0x%08x\n",
port_ptr->this_port.node_id);
i += scnprintf(buf + i, max - i, "Port_id: 0x%08x\n",
port_ptr->this_port.port_id);
i += scnprintf(buf + i, max - i, "# pkts tx'd %d\n",
port_ptr->num_tx);
i += scnprintf(buf + i, max - i, "# pkts rx'd %d\n",
port_ptr->num_rx);
i += scnprintf(buf + i, max - i, "# bytes tx'd %ld\n",
port_ptr->num_tx_bytes);
i += scnprintf(buf + i, max - i, "# bytes rx'd %ld\n",
port_ptr->num_rx_bytes);
spin_unlock_irqrestore(&port_ptr->port_lock, flags);
i += scnprintf(buf + i, max - i, "\n");
}
}
up_read(&local_ports_lock_lha2);
return i;
}
#define DEBUG_BUFMAX 4096
static char debug_buffer[DEBUG_BUFMAX];
static ssize_t debug_read(struct file *file, char __user *buf,
size_t count, loff_t *ppos)
{
int (*fill)(char *buf, int max) = file->private_data;
int bsize = fill(debug_buffer, DEBUG_BUFMAX);
return simple_read_from_buffer(buf, count, ppos, debug_buffer, bsize);
}
static int debug_open(struct inode *inode, struct file *file)
{
file->private_data = inode->i_private;
return 0;
}
static const struct file_operations debug_ops = {
.read = debug_read,
.open = debug_open,
};
static void debug_create(const char *name, mode_t mode,
struct dentry *dent,
int (*fill)(char *buf, int max))
{
debugfs_create_file(name, mode, dent, fill, &debug_ops);
}
static void debugfs_init(void)
{
struct dentry *dent;
dent = debugfs_create_dir("msm_ipc_router", 0);
if (IS_ERR(dent))
return;
debug_create("dump_local_ports", 0444, dent,
dump_local_ports);
debug_create("dump_remote_ports", 0444, dent,
dump_remote_ports);
debug_create("dump_control_ports", 0444, dent,
dump_control_ports);
debug_create("dump_servers", 0444, dent,
dump_servers);
debug_create("dump_xprt_info", 0444, dent,
dump_xprt_info);
debug_create("dump_routing_table", 0444, dent,
dump_routing_table);
}
#else
static void debugfs_init(void) {}
#endif
static int msm_ipc_router_add_xprt(struct msm_ipc_router_xprt *xprt)
{
struct msm_ipc_router_xprt_info *xprt_info;
struct msm_ipc_routing_table_entry *rt_entry;
xprt_info = kmalloc(sizeof(struct msm_ipc_router_xprt_info),
GFP_KERNEL);
if (!xprt_info)
return -ENOMEM;
xprt_info->xprt = xprt;
xprt_info->initialized = 0;
xprt_info->remote_node_id = -1;
INIT_LIST_HEAD(&xprt_info->pkt_list);
mutex_init(&xprt_info->rx_lock_lhb2);
mutex_init(&xprt_info->tx_lock_lhb2);
wake_lock_init(&xprt_info->wakelock,
WAKE_LOCK_SUSPEND, xprt->name);
xprt_info->need_len = 0;
xprt_info->abort_data_read = 0;
INIT_WORK(&xprt_info->read_data, do_read_data);
INIT_LIST_HEAD(&xprt_info->list);
xprt_info->workqueue = create_singlethread_workqueue(xprt->name);
if (!xprt_info->workqueue) {
kfree(xprt_info);
return -ENOMEM;
}
if (!strcmp(xprt->name, "msm_ipc_router_loopback_xprt")) {
xprt_info->remote_node_id = IPC_ROUTER_NID_LOCAL;
xprt_info->initialized = 1;
}
down_write(&xprt_info_list_lock_lha5);
list_add_tail(&xprt_info->list, &xprt_info_list);
up_write(&xprt_info_list_lock_lha5);
down_write(&routing_table_lock_lha3);
if (!routing_table_inited) {
init_routing_table();
rt_entry = alloc_routing_table_entry(IPC_ROUTER_NID_LOCAL);
add_routing_table_entry(rt_entry);
routing_table_inited = 1;
}
up_write(&routing_table_lock_lha3);
xprt->priv = xprt_info;
return 0;
}
static void msm_ipc_router_remove_xprt(struct msm_ipc_router_xprt *xprt)
{
struct msm_ipc_router_xprt_info *xprt_info;
if (xprt && xprt->priv) {
xprt_info = xprt->priv;
mutex_lock(&xprt_info->rx_lock_lhb2);
xprt_info->abort_data_read = 1;
mutex_unlock(&xprt_info->rx_lock_lhb2);
down_write(&xprt_info_list_lock_lha5);
list_del(&xprt_info->list);
up_write(&xprt_info_list_lock_lha5);
flush_workqueue(xprt_info->workqueue);
destroy_workqueue(xprt_info->workqueue);
wake_lock_destroy(&xprt_info->wakelock);
xprt->priv = 0;
kfree(xprt_info);
}
}
struct msm_ipc_router_xprt_work {
struct msm_ipc_router_xprt *xprt;
struct work_struct work;
};
static void xprt_open_worker(struct work_struct *work)
{
struct msm_ipc_router_xprt_work *xprt_work =
container_of(work, struct msm_ipc_router_xprt_work, work);
msm_ipc_router_add_xprt(xprt_work->xprt);
kfree(xprt_work);
}
static void xprt_close_worker(struct work_struct *work)
{
struct msm_ipc_router_xprt_work *xprt_work =
container_of(work, struct msm_ipc_router_xprt_work, work);
msm_ipc_cleanup_routing_table(xprt_work->xprt->priv);
msm_ipc_router_remove_xprt(xprt_work->xprt);
xprt_work->xprt->sft_close_done(xprt_work->xprt);
kfree(xprt_work);
}
void msm_ipc_router_xprt_notify(struct msm_ipc_router_xprt *xprt,
unsigned event,
void *data)
{
struct msm_ipc_router_xprt_info *xprt_info = xprt->priv;
struct msm_ipc_router_xprt_work *xprt_work;
struct rr_packet *pkt;
unsigned long ret;
if (!msm_ipc_router_workqueue) {
ret = wait_for_completion_timeout(&msm_ipc_local_router_up,
IPC_ROUTER_INIT_TIMEOUT);
if (!ret || !msm_ipc_router_workqueue) {
pr_err("%s: IPC Router not initialized\n", __func__);
return;
}
}
switch (event) {
case IPC_ROUTER_XPRT_EVENT_OPEN:
D("open event for '%s'\n", xprt->name);
xprt_work = kmalloc(sizeof(struct msm_ipc_router_xprt_work),
GFP_ATOMIC);
if (xprt_work) {
xprt_work->xprt = xprt;
INIT_WORK(&xprt_work->work, xprt_open_worker);
queue_work(msm_ipc_router_workqueue, &xprt_work->work);
} else {
pr_err("%s: malloc failure - Couldn't notify OPEN event",
__func__);
}
break;
case IPC_ROUTER_XPRT_EVENT_CLOSE:
D("close event for '%s'\n", xprt->name);
xprt_work = kmalloc(sizeof(struct msm_ipc_router_xprt_work),
GFP_ATOMIC);
if (xprt_work) {
xprt_work->xprt = xprt;
INIT_WORK(&xprt_work->work, xprt_close_worker);
queue_work(msm_ipc_router_workqueue, &xprt_work->work);
} else {
pr_err("%s: malloc failure - Couldn't notify CLOSE event",
__func__);
}
break;
}
if (!data)
return;
while (!xprt_info) {
msleep(100);
xprt_info = xprt->priv;
}
pkt = clone_pkt((struct rr_packet *)data);
if (!pkt)
return;
mutex_lock(&xprt_info->rx_lock_lhb2);
list_add_tail(&pkt->list, &xprt_info->pkt_list);
wake_lock(&xprt_info->wakelock);
mutex_unlock(&xprt_info->rx_lock_lhb2);
queue_work(xprt_info->workqueue, &xprt_info->read_data);
}
static int __init msm_ipc_router_init(void)
{
int i, ret;
struct msm_ipc_routing_table_entry *rt_entry;
msm_ipc_router_debug_mask |= SMEM_LOG;
ipc_rtr_log_ctxt = ipc_log_context_create(IPC_RTR_LOG_PAGES,
"ipc_router", 0);
if (!ipc_rtr_log_ctxt)
pr_err("%s: Unable to create IPC logging for IPC RTR",
__func__);
msm_ipc_router_workqueue =
create_singlethread_workqueue("msm_ipc_router");
if (!msm_ipc_router_workqueue)
return -ENOMEM;
debugfs_init();
for (i = 0; i < SRV_HASH_SIZE; i++)
INIT_LIST_HEAD(&server_list[i]);
for (i = 0; i < LP_HASH_SIZE; i++)
INIT_LIST_HEAD(&local_ports[i]);
down_write(&routing_table_lock_lha3);
if (!routing_table_inited) {
init_routing_table();
rt_entry = alloc_routing_table_entry(IPC_ROUTER_NID_LOCAL);
add_routing_table_entry(rt_entry);
routing_table_inited = 1;
}
up_write(&routing_table_lock_lha3);
ret = msm_ipc_router_init_sockets();
if (ret < 0)
pr_err("%s: Init sockets failed\n", __func__);
ret = msm_ipc_router_security_init();
if (ret < 0)
pr_err("%s: Security Init failed\n", __func__);
complete_all(&msm_ipc_local_router_up);
return ret;
}
module_init(msm_ipc_router_init);
MODULE_DESCRIPTION("MSM IPC Router");
MODULE_LICENSE("GPL v2");