Re: [PATCH 1/7] char/mei: PCI device and char driver support.

[Arnd Bergmann <arnd@arndb.de>]

On Tuesday 22 March 2011, Oren Weil wrote:
> +static ssize_t mei_read(struct file *file, char __user *ubuf,
> +                        size_t length, loff_t *offset)
> +{
> +       int i;
> +       int rets;
> +       int err;
> +       int if_num = iminor(file->f_dentry->d_inode);
> +       struct mei_file_private *file_ext = file->private_data;
> +       struct mei_cb_private *priv_cb_pos = NULL;
> +       struct mei_cb_private *priv_cb = NULL;
> +       struct mei_device *dev;
> +
> +       if (!mei_device)
> +               return -ENODEV;
> +
> +       dev = pci_get_drvdata(mei_device);
> +       if (if_num != MEI_MINOR_NUMBER || !dev || !file_ext)
> +               return -ENODEV;
> +

I'm very confused by this code and how you support multiple
instances of PCI devices, cb_list entries and concurrent connections:

* You look up a global pointer to mei_device, when there could
  be multiple PCI devices.

* You use the minor number of the chardev as an index throughout
  the code, but bail out if it does not match a constant.

* You walk a list of mei_cb_private entries when you should have
  just a single one for this file. In other places, you walk a list
  of mei_file_private entries.

I think you should try to document your data structures and how
they relate to each other, and remove all that turn out not to
be needed in the process.

>From the brief look I had at this rather complex code, I would
assume that it could look something like this:

* In the PCI probe function, create a misc character device for
  every instance you find behind each device. Remove the global
  variables and constants.

* Unify mei_file_private and mei_cb_private, allocate at open
  time but fill when the ioctl gets called.

Would that work, or am I missing something major?

	Arnd