Newbie: Access stack avriables in a kernel module

Newbie: Access stack avriables in a kernel module

[Wolfram Gettert]

Hi all,
I am quite new to Systemtap. I have read the tutorial. Now I am trying to
do some examples on my own. I like to access some variables inside a
kernel module.

I understand that with $xxx in the probe I can access a variable xxx in the
kernel module. I have managed that for static int variable of the module.

Now, I want to a access a parameter in a function of the kernel module.
I created the following probe:

probe module("mplex").function("mplex_fop_ioctl").return
{
  printf("Call %s arg=%u, res=%i, return=%i\n", probefunc(),$arg, $res,
$return)
}

The code of the function in the kernel is:

static int mplex_fop_ioctl(struct inode *inode, struct file *filp,
                           unsigned int cmd, unsigned long arg)
{
  int res=0;

  struct mplex_dev *mplex=filp->private_data;

  PRINTK_DEBUG("entering mplex_fop_ioctl(), minor=%i\n",
MINOR(mplex->cdev.dev));

  /* Check that there is no IOCTL() confict */
  if ((_IOC_TYPE(cmd) != MPLEX_IOC_MAGIC) || (_IOC_NR(cmd) >
MPLEX_IOC_MAXNR)) {
    PRINTK_ERR("inappropriate ioctl() for device\n");
    return -ENOTTY;
  }
  PRINTK_INFO("cmd: %i", cmd);
  switch (cmd)
  {
    case MIOQPRIV1:
      PRINTK_INFO("*mp: %p\n", mplex);
      PRINTK_INFO("minor: %u, priv1=%u\n", MINOR(mplex->cdev.dev),
mplex->priv1);
      return mplex->priv1;
      break;

    case MIOTPRIV1:
      mplex->priv1=arg;
      break;

    default:
        return -ENOTTY;
  }

  return res;
}


The ouput I get from stap is:

Call mplex_fop_ioctl arg=10, res=96, return=0
Call mplex_fop_ioctl arg=20, res=96, return=0
Call mplex_fop_ioctl arg=30, res=96, return=0
Call mplex_fop_ioctl arg=40, res=96, return=0
Call mplex_fop_ioctl arg=40, res=96, return=10
Call mplex_fop_ioctl arg=40, res=96, return=20
Call mplex_fop_ioctl arg=40, res=96, return=30
Call mplex_fop_ioctl arg=40, res=96, return=40

The values of arg and return are Ok. But as you see in the code of the
function above, res is initialized with 0. But the probe prints another
value. Why?

Is it bug? Have I missunderstood something?

Thanks for any comment on that.

I am using
SystemTap translator/driver (version 1.0/0.143 Debian version 1.0-2)
on a x86 2.6.32-24-generic #41-Ubuntu SMP.


Wolfram

Re: Newbie: Access stack avriables in a kernel module

[Josh Stone]

Hi-

On 08/25/2010 07:40 AM, Wolfram Gettert wrote:
> probe module("mplex").function("mplex_fop_ioctl").return
> {
>   printf("Call %s arg=%u, res=%i, return=%i\n", probefunc(),$arg, $res,
> $return)
> }

First, I hope you're aware that accessing variables in a .return probe
means you get the value from the *entry* time of the function.  That
might be interesting for $arg, but I think $res won't be helpful.

> The code of the function in the kernel is:
> 
> static int mplex_fop_ioctl(struct inode *inode, struct file *filp,
>                            unsigned int cmd, unsigned long arg)
> {
>   int res=0;
[...]
> The values of arg and return are Ok. But as you see in the code of the
> function above, res is initialized with 0. But the probe prints another
> value. Why?

res=0 is the first thing the function does, but the probe is placed
before the function has done anything at all.  So the value you're
reading is probably whatever happened to be in that location already,
whether it's a register or on the stack.


Josh

Re: Newbie: Access stack avriables in a kernel module

[Wolfram Gettert (News)]

Hi Josh,
thanks for the quick response.

> First, I hope you're aware that accessing variables in a .return probe
> means you get the value from the *entry* time of the function.  That
> might be interesting for $arg, but I think $res won't be helpful.
Understood, I was not aware about that.

>
>> The code of the function in the kernel is:
>>
>> static int mplex_fop_ioctl(struct inode *inode, struct file *filp,
>>                            unsigned int cmd, unsigned long arg)
>> {
>>   int res=0;
> [...]
>> The values of arg and return are Ok. But as you see in the code of the
>> function above, res is initialized with 0. But the probe prints another
>> value. Why?
>
> res=0 is the first thing the function does, but the probe is placed
> before the function has done anything at all.  So the value you're
> reading is probably whatever happened to be in that location already,
> whether it's a register or on the stack.
>
Understood.


Thanks a lot.

Wolfram

--