How to Profile Memory in Linux

/proc/meminfo

Video4Linux camera driver分析

/proc/meminfo

How to Profile Memory in Linux

2010-12-10 10:46:42| 分类：

技术 | 标签：memory process swap unit graph
|举报
|字号大中小 订阅

http://www.freshblurbs.com/how-profile-memory-linux
1. Introduction
It's important to determine how your system utilizes it's

resources. If your systems performance is unacceptable, it is

necessary to determine which resource is slowing the system

down. This document attempts to identify the following:
a. What is the system memory usage per unit time?

b. How much swap is being used per unit time?

c. What does each process' memory use look like over time?

d. What processes are using the most memory?
I used a RedHat-7.3 machine (kernel-2.4.18) for my experiments,

but any modern Linux distribution with the commands "ps" and

"free" would work.
2. Definitions
RAM (Random Access Memory) - Location where programs reside when

they are running. Other names for this are system memory or

physical memory. The purpose of this document is to determine if

you have enough of this.
Memory Buffers - A page cache for the virtual memory system. The

kernel keeps track of frequently accessed memory and stores the

pages here.
Memory Cached - Any modern operating system will cache files

frequently accessed. You can see the effects of this with the

following commands:
for i in 1 2 ; do

free -o

time grep -r foo /usr/bin >/dev/null 2>/dev/null

done
Memory Used - Amount of RAM in use by the computer. The kernel

will attempt to use as much of this as possible through buffers

and caching.
Swap - It is possible to extend the memory space of the computer

by using the hard drive as memory. This is called swap. Hard

drives are typically several orders of magnitude slower than RAM

so swap is only used when no RAM is available.
Swap Used - Amount of swap space used by the computer.
PID (Process IDentifier) - Each process (or instance of a running

program) has a unique number. This number is called a PID.
PPID (Parent Process IDentifier) - A process (or running program)

can create new processes. The new process created is called a

child process. The original process is called the parent

process. The child process has a PPID equal to the PID of the

parent process. There are two exceptions to this rule. The first

is a program called "init". This process always has a PID of 1 and

a PPID of 0. The second exception is when a parent process exit

all of the child processes are adopted by the "init" process and

have a PPID of 1.
VSIZE (Virtual memory SIZE) - The amount of memory the process is

currently using. This includes the amount in RAM and the amount in

swap.
RSS (Resident Set Size) - The portion of a process that exists in

physical memory (RAM). The rest of the program exists in swap. If

the computer has not used swap, this number will be equal to

VSIZE.
3. What consumes System Memory?
The kernel - The kernel will consume a couple of MB of memory. The

memory that the kernel consumes can not be swapped out to

disk. This memory is not reported by commands such as "free" or

"ps".
Running programs - Programs that have been executed will consume

memory while they run.
Memory Buffers - The amount of memory used is managed by the

kernel. You can get the amount with "free".
Memory Cached - The amount of memory used is managed by the

kernel. You can get the amount with "free".
4. Determining System Memory Usage
The inputs to this section were obtained with the command:
free -o
The command "free" is a c program that reads the "/proc"

filesystem.
There are three elements that are useful when determining the

system memory usage. They are:
a. Memory Used

b. Memory Used - Memory Buffers - Memory Cached

c. Swap Used
A graph of "Memory Used" per unit time will show the "Memory Used"

asymptotically approach the total amount of memory in the system

under heavy use. This is normal, as RAM unused is RAM wasted.
A graph of "Memory Used - Memory Buffered - Memory Cached" per

unit time will give a good sense of the memory use of your

applications minus the effects of your operating system. As you

start new applications, this value should go up. As you quit

applications, this value should go down. If an application has a

severe memory leak, this line will have a positive slope.
A graph of "Swap Used" per unit time will display the swap

usage. When the system is low on RAM, a program called kswapd will

swap parts of process if they haven't been used for some time. If

the amount of swap continues to climb at a steady rate, you may

have a memory leak or you might need more RAM.
5. Per Process Memory Usage
The inputs to this section were obtained with the command:
ps -eo pid,ppid,rss,vsize,pcpu,pmem,cmd -ww --sort=pid
The command "ps" is a c program that reads the "/proc"

filesystem.
There are two elements that are useful when determining the per

process memory usage. They are:
a. RSS

b. VSIZE
A graph of RSS per unit time will show how much RAM the process is

using over time.
A graph of VSIZE per unit time will show how large the process is

over time.
6. Collecting Data
a. Reboot the system. This will reset your systems memory use
b. Run the following commands every ten seconds and redirect the

results to a file.
free -o

ps -eo pid,ppid,rss,vsize,pcpu,pmem,cmd -ww --sort=pid
c. Do whatever you normally do on your system
d. Stop logging your data
7. Generate a Graph
a. System Memory Use
For the output of "free", place the following on one graph
1. X-axis is "MB Used"
2. Y-axis is unit time
3. Memory Used per unit time
4. Memory Used - Memory Buffered - Memory Cached per unit time
5. Swap Used per unit time
b. Per Process Memory Use
For the output of "ps", place the following on one graph
1. X-axis is "MB Used"
2. Y-axis is unit time
3. For each process with %MEM > 10.0
a. RSS per unit time
b. VSIZE per unit time
8. Understand the Graphs
a. System Memory Use
"Memory Used" will approach "Memory Total"
If "Memory Used - Memory Buffered - Memory Cached" is 75% of

"Memory Used", you either have a memory leak or you need to

purchase more memory.
b. Per Process Memory Use
This graph will tell you what processes are hogging the

memory.
If the VSIZE of any of these programs has a constant, positive

slope, it may have a memory leak.

Video4Linux camera driver分析

2010-12-08 21:52:53| 分类：

技术 | 标签：irq iowait dma video4linux camera
|举报
|字号大中小 订阅

avengers_lite.c中调用pxa168_add_cam()来增加pxa168_device_camera设备。该设备在device.c中定义：
PXA168_DEVICE(camera, "pxa168-camera", -1, CI, 0xd420a000, 0xfff);
这里指定了IRQ为IRQ_PXA168_CI(即42)，寄存器起始地址为0xd420a000，范围为0xfff

pxa910_camera_probe
1. init_waitqueue_head(&cam->iowait), 初始化wait queue
2. tasklet_init(&cam->s_tasklet, ccic_frame_tasklet, (unsigned long) cam), 初始化tasklet
3. cam->regs = ioremap(res->start, SZ_4K), 讲寄存器地址映射到虚拟地址空间
4. request_irq(cam->irq, ccic_irq, ...), 申请irq。这里取得的irq即为IRQ_PXA168_CI
5. video_register_device(&cam->v4ldev, VFL_TYPE_GRABBER, -1), 注册video设备
6. ccic_alloc_dma_bufs，分配DMA buffer

开启DMA传输：
ccic_vidioc_streamon
ccic_read_setup
ccic_ctlr_configure(cam);
ccic_ctlr_dma /* V4L2_MEMORY_USERPTR */
ccic_ctlr_dma_mmap /* V4L2_MEMORY_MMAP方式 */
ccic_reg_write(cam, REG_Y0BAR, cam->dma_handles[0]); 配置DMA传输地址
ccic_ctlr_start /* 启动DMA传输 */
ccic_reg_set_bit(cam, REG_CTRL0, C0_ENABLE)

当irq发生时：
ccic_irq()
ccic_frame_irq()
ccic_frame_complete()
tasklet_schedule(&cam->s_tasklet); /* V4L2_MEMORY_MMAP方式 */
ccic_switch_dma(cam, frame); /* V4L2_MEMORY_USERPTR */

ccic_frame_tasklet
1. 从cam->dma_bufs中拷贝数据，添加到cam->sb_full
2. 如果cam->sb_full中有数据，wake_up(&cam->iowait)

ccic_vidioc_dqbuf阻塞在wait_event_interruptible(cam->iowait,!list_empty(&cam->sb_full))