linux 巨页使用测试以及勘误1

linux使用hugetlbfs的方式来支持巨页，也成为大页。

网上看到有人说巨页不支持read，和write调用，只支持mmap，但是看3.10内核代码的时候发现：

const struct file_operations hugetlbfs_file_operations = {
.read = hugetlbfs_read,--------------------------------初始化了。
.mmap = hugetlbfs_file_mmap,
.fsync = noop_fsync,
.get_unmapped_area = hugetlb_get_unmapped_area,
.llseek = default_llseek,
};

查看vfs_read代码，要确定下是否hugetlbfs_read就是file->f_op->read

ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
{
ssize_t ret;

if (!(file->f_mode & FMODE_READ))
return -EBADF;
if (!file->f_op || (!file->f_op->read && !file->f_op->aio_read))
return -EINVAL;
if (unlikely(!access_ok(VERIFY_WRITE, buf, count)))
return -EFAULT;

ret = rw_verify_area(READ, file, pos, count);
if (ret >= 0) {
count = ret;
if (file->f_op->read)
ret = file->f_op->read(file, buf, count, pos);

使用如下测试代码分别测试：

#include <fcntl.h>
#include <sys/mman.h>
#include <errno.h>
#include <unistd.h>
#include <stdio.h>

#define MAP_LENGTH (10*1024*1024)
#define TEST_PATH ("/mnt/huge_caq/test")
#define TEST_READ 1
#define TEST_WRITE 0
#define TEST_MMAP 1

int main()
{
int fd;
void * addr;
ssize_t ret=0;

/* create a file in hugetlb fs */
fd = open(TEST_PATH, O_CREAT | O_RDWR);
if(fd < 0){
perror("Err: ");
return -1;
}
#if 0
#if TEST_WRITE
ret = write(fd,"hello\r\n",7);
perror("Err: ");
#endif
#endif
#if TEST_READ
char buf[10];
ret = read(fd,buf,7);
printf("read ret=%d\r\n",ret);
#endif
#if TEST_MMAP
/* map the file into address space of current application process */
addr = mmap(0, MAP_LENGTH, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
if(addr == MAP_FAILED){
perror("Err: ");
close(fd);
unlink(TEST_PATH);
return -1;
}

/* from now on, you can store application data on huage pages via addr */
memcpy(addr,"hello\r\n",7);
#if TEST_WRITE
ret = write(fd,"hello\r\n",7);
perror("Err: ");
#endif

#if TEST_READ
memset(buf,0,10);
ret = read(fd,buf,7);
if(ret)
{
printf("read ret=%d,buf=%s\r\n",ret,buf);
}
#endif
munmap(addr, MAP_LENGTH);
#if TEST_READ
memset(buf,0,10);
ret = read(fd,buf,7);
printf("read ret=%d\r\n",ret);
#endif
#endif
close(fd);
unlink(TEST_PATH);
return 0;
}

然后gdb跟踪如下，发现是可以读的，读出来数据也是正常的。

54 memset(buf,0,10);
(gdb)
55 ret = read(fd,buf,7);
(gdb)
56 if(ret)
(gdb) p ret
$1 = 7
(gdb) p buf
$2 = "hello\r\n\000\000"

使用如下脚本确定file->f_op->read的对应函数

probe kernel.function("vfs_read").return
{
#print_backtrace();
if(($count==7)&&($file->f_op)&&($file->f_op->read))
{
printf("begin pid=%d,execname=%s,f_op=%s,read=%s,aio=%x\r\n",pid(),execname(),symname($file->f_op),symname($file->f_op->read),$file->f_op->aio_read);
}
}

begin pid=30371,execname=main.o,f_op=hugetlbfs_file_operations,read=hugetlbfs_read,aio=0，没错，就是hugetlbfs_read。

从测试结果看，是支持read的。至于从哪个内核版本支持的，没有继续研究，有兴趣的同学可以跟一下git的记录。