msm平台，AP和CP封装在一起，公用一块内存。所以AP需要负责把整个modem, TZ , rpm等binary拷贝到内存中以供modem等subsystem去运行。那AP这边是怎么分配这些内存，又是怎么读出来相关的binary，又如何把binary上传上去的呢？？
相关的feature
CONFIG_FW_LOADER
CONFIG_FW_LOADER_USER_HELPER
modem使用的内存申请
要设置modem的内存大小，必须首先需要确认modem binary的大小，modem需要使用的内存大小等。这个在CMA相关的内容中说过。这里在说一下高通msm8916平台，modem大小检查以及修改方法。
1) modem binary的大小可以从以下编译的log里边看出来！！(modem_proc/build/ms目录下的pplk-XXX.log或者build_xxxx.log)。
根据大小对齐1MB大小，就是modem binary需要流出来的大小。看如下例子里边的log，总的大小是77.04，
所以需要在上面的dtsi文件中留出来78MB就可以。
Image loaded at virtual address 0xc0000000
Image: 55.44 MiB
AMSS Heap: 7.50 MiB (dynamic)
MPSS Heap: 4.00 MiB (dynamic)
DSM Pools: 5.06 MiB
Q6Zip RO, Swap Pool: 2.00 MiB (dynamic)
Q6Zip RW, Swap Pool: 1.00 MiB (dynamic)
Q6Zip RW, dlpager Heap: 1.00 MiB
Extra: 0.54 MiB
Pad ding: 0.37 MiB
End Address Alignment: 0.13 MiB
Total: 77.04 MiB
Available: 7.96 MiB
2) 然后去修改modem_proc/config/xxx/ 目录下的cust_config.xml文件中修改modem大小
以下是modem相关的device tree的设置。这些内容也在CMA和ion内存相关的帖子里边都讲过。
但之前有一个疑问就是，在CMA预留了一段内存之后，会把这个赋值给modem的dev->cma_area，然后在分配需要使用的内存的时候从dev->cma_area中取出来，那这个过程好像跟ion内存没有什么关系。能不能去掉下面msmxxx-ion.dtsi中
modem_adsp_mem相关的设置呢？？
是可以的！！！其他几个DMA区域，如果直接从CMA分配的话，应该都可以从msmxxx-ion.dtsi文件中去掉！！
也就是说下面qcom,ion-heap-type = “DMA”的部分其实都可以从msm8916-ion.dtsi文件中去掉，不影响。
//modem相关内存的device tree设置
//pil设备相关的device tree定义
qcom,mss@4080000 {
compatible = "qcom,pil-q6v56-mss";
....
linux,contiguous-region = <&modem_adsp_mem>;
};
//msmxxx-ion.dtsi定义了如下，上面说了这个部分其实是可以去掉的，不会影响相关内存的分配！！
qcom,ion-heap@26 { /* MODEM HEAP */
compatible = "qcom,msm-ion-reserve";
reg = <26>;
linux,contiguous-region = <&modem_adsp_mem>;
qcom,ion-heap-type = "DMA";
};
//msmxxx-memory.dtsi定义了如下内容
modem_adsp_mem: modem_adsp_region@0 {
linux,reserve-contiguous-region;
linux,reserve-region;
linux,remove-completely;
reg = <0x0 0x86800000 0x0 0x05800000>;
label = "modem_adsp_mem";
};
modem_adsp_mem指定的区域，需要分配出来，以供下载modem binary。
//pil_mss_driver_probe()->pil_subsys_init()
static int pil_subsys_init(struct modem_data *drv,
struct platform_device *pdev)
{
...
drv->subsys_desc.name = "modem";
drv->subsys_desc.dev = &pdev->dev;
drv->subsys_desc.owner = THIS_MODULE;
drv->subsys_desc.shutdown = modem_shutdown;
drv->subsys_desc.powerup = modem_powerup;
drv->subsys_desc.ramdump = modem_ramdump;
drv->subsys_desc.free_memory = modem_free_memory;
drv->subsys_desc.crash_shutdown = modem_crash_shutdown;
drv->subsys_desc.err_fatal_handler = modem_err_fatal_intr_handler;
drv->subsys_desc.stop_ack_handler = modem_stop_ack_intr_handler;
drv->subsys_desc.wdog_bite_handler = modem_wdog_bite_intr_handler;
drv->subsys = subsys_register(&drv->subsys_desc);
drv->ramdump_dev = create_ramdump_device("modem", &pdev->dev);
...
return ret;
}
之后在modem_powerup()的时候,会先根据modem binary的elf结构独处modem的大小等，然后计算出align之后应该的大小。
pil_boot()-> request_firmware()读出elf头并计算大小等。
在pil_setup_region()->pil_alloc_region()的时候，传进去的大小就是从上面读出来的大小。
pil_alloc_region min_addr = 0xc0000000 , max_addr = 0xc2b00000 , aligned_size = 0x2b00000
1
这里看着和实际的内存大小一致！！ 可能是因为留出来的CMA区域的大小正好和这个大小一致才这样的。
在实际调试过程中，也可以打印这个大小之后，调整CMA大小。
再看看实际的CMA大小是怎么申请的。
//调用顺序
pil_boot()->pil_init_mmap()->pil_setup_region()->pil_alloc_region()->
dma_alloc_attrs()->arm_dma_alloc()->__dma_alloc()->__alloc_from_contiguous()->
1
2
3
这个调用的顺序，一步一步往下看可以看到，实际上分配的区域是一块CMA区域，而且就是在CMA注册之后，在相应的platform设备注册的时候保存到dev->cma_area中的区域。
在相应的设备注册的时候，如果设备的device tree中有”linux,contiguous-region”的时候，就会寻找相应的CMA区域并进行保留。这都是因为注册了platform_bus_typ的notifier函数
bus_register_notifier(&platform_bus_type, &cma_dev_init_nb);
1
看下面的log。
<6>[0.487642] [0:swapper/0:1] cma: Assigned CMA region at 0 to 1de0000.qcom,venus device
<6>[0.489469] [0:swapper/0:1] cma: Assigned CMA region at 0 to 4080000.qcom,mss device
<6>[0.490756] [0:swapper/0:1] cma: Assigned CMA region at 0 to a21b000.qcom,pronto device
<6>[1.125342] [0:swapper/0:1] cma: Assigned CMA region at 0 to 8.qcom,ion-heap device
<6>[1.125793] [0:swapper/0:1] cma: Assigned CMA region at 0 to 1b.qcom,ion-heap device
<6>[1.126233] [0:swapper/0:1] cma: Assigned CMA region at 0 to 1c.qcom,ion-heap device
<6>[1.126671] [0:swapper/0:1] cma: Assigned CMA region at 0 to 17.qcom,ion-heap device
<6>[1.127298] [0:swapper/0:1] cma: Assigned CMA region at 0 to 1a.qcom,ion-heap device
这里看到4080000.qcom,mss这个device相应的区域已经保留了CMA区域。
然后在上面进行分配的时候，在
__alloc_from_contiguous()->dma_alloc_from_contiguous()->dev_get_cma_area()函数中取到
相应的dev->cma_area。
modem相关内存的使用和下载
pil_load_seg()->request_firmware_direct()->_request_firmware()函数身生成相应的device节点，并通知ueventd去读取相应的binary然后下载。以下是pil_load_seg里边打印的正在试图下载的binary。
<6>[29.129737] [1:init:1] pil_load_seg fw_name = modem.b02
<6>[29.157808] [1:init:1] pil_load_seg fw_name = modem.b07
<6>[29.191477] [1:init:1] pil_load_seg fw_name = modem.b17
<6>[29.348480] [1:init:1] pil_load_seg fw_name = modem.b19
<6>[29.409733] [1:init:1] pil_load_seg fw_name = modem.b20
<6>[29.489639] [1:init:1] pil_load_seg fw_name = modem.b23
<6>[29.519624] [1:init:1] pil_load_seg fw_name = modem.b24
<6>[29.549829] [1:init:1] pil_load_seg fw_name = modem.b25
<6>[29.591918] [1:init:1] pil_load_seg fw_name = modem.b27
<6>[31.997036] [0:wcnss_service:307] pil_load_seg fw_name = wcnss.b02
<6>[32.658390] [0:wcnss_service:307] pil_load_seg fw_name = wcnss.b04
<6>[32.693754] [0:wcnss_service:307] pil_load_seg fw_name = wcnss.b06
<6>[32.848104] [3:wcnss_service:307] pil_load_seg fw_name = wcnss.b09
<6>[32.854061] [3:wcnss_service:307] pil_load_seg fw_name = wcnss.b10
<6>[32.876115] [3:wcnss_service:307] pil_load_seg fw_name = wcnss.b11
<6>[37.384287] [1:TimedEventQueue:771] pil_load_seg fw_name = venus.b02
<6>[37.438222] [1:TimedEventQueue:771] pil_load_seg fw_name = venus.b03
<6>[37.484909] [2:TimedEventQueue:771] pil_load_seg fw_name = venus.b041
在 _request_firmware()->fw_load_from_user_helper()->_request_firmware_load()函数中就在生成相应的dev节点，并通知ueventd。
static int _request_firmware_load(struct firmware_priv *fw_priv, bool uevent,
long timeout)
{
int retval = 0;
struct device *f_dev = &fw_priv->dev;
struct firmware_buf *buf = fw_priv->buf;
struct bin_attribute *fw_attr_data = buf->dest_addr ?
&firmware_direct_attr_data : &firmware_attr_data;
/* fall back on userspace loading */
buf->is_paged_buf = buf->dest_addr ? false : true;
dev_set_uevent_suppress(f_dev, true);
/* Need to pin this module until class device is destroyed */
__module_get(THIS_MODULE);
retval = device_add(f_dev);
//以下生成的data和loading节点，用于ueventd读取相应的binary，然后通过节点加载到内存的。
//用于下载的节点，
retval = device_create_bin_file(f_dev, fw_attr_data);
//生成一个loading的节点，loading节点用于控制的
retval = device_create_file(f_dev, &dev_attr_loading);
if (retval) {
dev_err(f_dev, "%s: device_create_file failed\n", __func__);
goto err_del_bin_attr;
}
if (uevent) { //这里正在通知ueventd
dev_set_uevent_suppress(f_dev, false);
dev_dbg(f_dev, "firmware: requesting %s\n", buf->fw_id);
if (timeout != MAX_SCHEDULE_TIMEOUT)
schedule_delayed_work(&fw_priv->timeout_work, timeout);
kobject_uevent(&fw_priv->dev.kobj, KOBJ_ADD);
}
wait_for_completion(&buf->completion);
cancel_delayed_work_sync(&fw_priv->timeout_work);
}
_request_firmware() -> assign_firmware_buf() 这是做什么的？？
来看一下ueventd.c文件中是怎么检测这个然后读binary，通过loading节点加载binary的。
int ueventd_main(int argc, char **argv){
...
while(1) {
ufd.revents = 0;
nr = poll(&ufd, 1, -1);
if (nr <= 0)
continue;
if (ufd.revents & POLLIN)
handle_device_fd();
}
}
void handle_device_fd(){
...
handle_firmware_event(&uevent);//process_firmware_event()
}
#define SYSFS_PREFIX "/sys"
#define FIRMWARE_DIR1 "/etc/firmware"
#define FIRMWARE_DIR2 "/vendor/firmware"
#define FIRMWARE_DIR3 "/firmware/image"
#define FIRMWARE_DIR4 "/firmware-modem/image"
#define DEVICES_BASE "/devices/soc.0"
static void process_firmware_event(struct uevent *uevent){
...
l = asprintf(&root, SYSFS_PREFIX"%s/", uevent->path);
l = asprintf(&loading, "%sloading", root);
l = asprintf(&file1, FIRMWARE_DIR1"/%s", uevent->firmware);
l = asprintf(&file2, FIRMWARE_DIR2"/%s", uevent->firmware);
l = asprintf(&file3, FIRMWARE_DIR3"/%s", uevent->firmware);
l = asprintf(&file4, FIRMWARE_DIR4"/%s", uevent->firmware);
loading_fd = open(loading, O_WRONLY);
...
if(!load_firmware(fw_fd, loading_fd, data_fd)) //加载binary
INFO("firmware: copy success { '%s', '%s' }\n", root, uevent->firmware);
else
INFO("firmware: copy failure { '%s', '%s' }\n", root, uevent->firmware);
}
以下看看ueventd中，真正把读到的binary，传给kernel的函数
static int load_firmware(int fw_fd, int loading_fd, int data_fd)
{
struct stat st;
long len_to_copy;
int ret = 0;
//fstat查看binary的信息，读出来size等
if(fstat(fw_fd, &st) < 0)
return -1;
len_to_copy = st.st_size;
write(loading_fd, "1", 1); /* start transfer */
while (len_to_copy > 0) {
char buf[PAGE_SIZE];
ssize_t nr;
//读
nr = read(fw_fd, buf, sizeof(buf));
if(!nr)
break;
if(nr < 0) {
ret = -1;
break;
}
len_to_copy -= nr;
while (nr > 0) {
ssize_t nw = 0;
//写到data节点
nw = write(data_fd, buf + nw, nr);
if(nw <= 0) {
ret = -1;
goto out;
}
nr -= nw;
}
}
out:
if(!ret) //loading节点用于通知kernel加载情况！！
write(loading_fd, "0", 1); /* successful end of transfer */
else
write(loading_fd, "-1", 2); /* abort transfer */
return ret;
}
1
2
3
4
5
6
内核中，data节点出来write的函数在_request_firmware_load()中根据buf->dest_addr的值有所不同
static int _request_firmware_load(struct firmware_priv *fw_priv, bool uevent,
long timeout)
{
...
struct bin_attribute *fw_attr_data = buf->dest_addr ?
&firmware_direct_attr_data : &firmware_attr_data;
...
}
在下载modem.bxx的时候应该都是有buf->dest_addr才对
<6>[29.355860] [0:init:1] pil_load_seg fw_name = modem.b02
<6>[29.361829] [0:init:1] fw_load_from_user_helper start
<6>[29.368051] [0:init:1] _request_firmware_load buf->dest_addr = 0x86800000
<6>[29.380308] [0:ueventd:230] firmware_loading_store started
...
<6>[29.391942] [0:init:1] pil_load_seg fw_name = modem.b07
<6>[29.398801] [0:init:1] fw_load_from_user_helper start
<6>[29.404996] [0:init: 1] _request_firmware_load buf->dest_addr = 0x86840000
...
//write(data_fd, buf + nw, nr); buf对应buffer？ offset? count对应nr??
static ssize_t firmware_direct_write(struct file *filp, struct kobject *kobj,
struct bin_attribute *bin_attr,
char *buffer, loff_t offset, size_t count)
{
struct device *dev = kobj_to_dev(kobj);
struct firmware_priv *fw_priv = to_firmware_priv(dev);
//获取uevent读取modem binary时候读到的内容，保存在firmware_priv中。firmware_priv中的firmware_buf保存了binary的物理地址，大小等等信息
struct firmware *fw;
ssize_t retval;
if (!capable(CAP_SYS_RAWIO))
return -EPERM;
mutex_lock(&fw_lock);
fw = fw_priv->fw;
if (!fw || test_bit(FW_STATUS_DONE, &fw_priv->buf->status)) {
retval = -ENODEV;
goto out;
}
retval = __firmware_data_rw(fw_priv, buffer, &offset, count, 0);
if (retval < 0)
goto out;
fw_priv->buf->size = max_t(size_t, offset, fw_priv->buf->size);
out:
mutex_unlock(&fw_lock);
return retval;
}
static int __firmware_data_rw(struct firmware_priv *fw_priv, char *buffer,
loff_t *offset, size_t count, int read)
{
u8 __iomem *fw_buf;
struct firmware_buf *buf = fw_priv->buf;
int retval = count;
if ((*offset + count) > buf->dest_size) {
pr_debug("%s: Failed size check.\n", __func__);
retval = -EINVAL;
goto out;
}
//fw_buf 就是要拷贝到内存中的modem binary物理地址对应的虚拟地址。
//map_fw_mem函数中，会根据虚拟地址以及需要拷贝的大小，map出一段虚拟地址。
//map一段物理地址，然后返回内核可以访问的虚拟地址，，这个是通过ioremap相关的函数实现的
fw_buf = buf->map_fw_mem(buf->dest_addr + *offset, count,
buf->map_data);
if (!fw_buf) {
pr_debug("%s: Failed ioremap.\n", __func__);
retval = -ENOMEM;
goto out;
}
//读写，直接拷贝就可以
if (read)
memcpy(buffer, fw_buf, count);
else
memcpy(fw_buf, buffer, count);
*offset += count;
buf->unmap_fw_mem(fw_buf, count, buf->map_data);
out:
return retval;
}
static void *map_fw_mem(phys_addr_t paddr, size_t size, void *data)
{
struct pil_map_fw_info *info = data;
return dma_remap(info->dev, info->region, paddr, size,
&info->attrs);
}
static inline void *dma_remap(struct device *dev, void *cpu_addr,
dma_addr_t dma_handle, size_t size, struct dma_attrs *attrs)
{
const struct dma_map_ops *ops = get_dma_ops(dev);
BUG_ON(!ops);
if (!ops->remap) {
WARN_ONCE(1, "Remap function not implemented for %pS\n",
ops->remap);
return NULL;
}
return ops->remap(dev, cpu_addr, dma_handle, size, attrs);
}
static void *arm_dma_remap(struct device *dev, void *cpu_addr,
dma_addr_t handle, size_t size,
struct dma_attrs *attrs)
{
struct page *page = pfn_to_page(dma_to_pfn(dev, handle));
pgprot_t prot = __get_dma_pgprot(attrs, PAGE_KERNEL);
unsigned long offset = handle & ~PAGE_MASK;
size = PAGE_ALIGN(size + offset);
return __dma_alloc_remap(page, size, GFP_KERNEL, prot,
__builtin_return_address(0)) + offset;
}
static void *
__dma_alloc_remap(struct page *page, size_t size, gfp_t gfp, pgprot_t prot,
const void *caller)
{
struct vm_struct *area;
unsigned long addr;
/*
* DMA allocation can be mapped to user space, so lets
* set VM_USERMAP flags too.
*/
//得到一段满足要求的vm_struct。这里
area = get_vm_area_caller(size, VM_ARM_DMA_CONSISTENT | VM_USERMAP,
caller);
if (!area)
return NULL;
addr = (unsigned long)area->addr;
area->phys_addr = __pfn_to_phys(page_to_pfn(page));
//addr是得到的vm_struct对应的虚拟地址，内核可以访问的
//所以根据物理地址以及对应的虚拟地址以及大小等情况，ioremap_page_range会做一个page table
//这样内核就可以直接访问这段内存
if (ioremap_page_range(addr, addr + size, area->phys_addr, prot)) {
vunmap((void *)addr);
return NULL;
}
return (void *)addr;//返回虚拟内存，现在这个虚拟内存就可以直接访问了
}
和ioremap_page_range()比较像。
