On 10/08/11 10:15, Emilio G. Cota wrote:
> On Wed, Aug 10, 2011 at 08:39:07 +0100, Martyn Welch wrote:
>> And I think you need to go and do a grep of the code and find out where those
>> functions are actually used, rather than blindly relying on the comment.
> (snip)
>> Go grep the code.
>
> /me greps once again..
>
> RapidIO: there are no rapidIO drivers upstream, only switches
> and rionet, which does not manage RapidIO devices (it just sends
> Ethernet packets on top of RapidIO's messaging). So obviously
> there aren't any callers.
>
Yes there are. In rio-driver.c for a start:
/**
* rio_device_probe - Tell if a RIO device structure has a matching RIO
device id structure
* @dev: the RIO device structure to match against
*
* return 0 and set rio_dev->driver when drv claims rio_dev, else error
*/
static int rio_device_probe(struct device *dev)
{
struct rio_driver *rdrv = to_rio_driver(dev->driver);
struct rio_dev *rdev = to_rio_dev(dev);
int error = -ENODEV;
const struct rio_device_id *id;
if (!rdev->driver && rdrv->probe) {
if (!rdrv->id_table)
return error;
id = rio_match_device(rdrv->id_table, rdev);
rio_dev_get(rdev);
if (id)
error = rdrv->probe(rdev, id);
if (error >= 0) {
rdev->driver = rdrv;
error = 0;
} else
rio_dev_put(rdev);
}
return error;
}
Doing what Manohar suggested and I have agreed with.
> PCI: pci_dev_get() referenced in 60 files. Another way of
> explicitly incrementing the refcount of a pci device is with
> pci_get_device(), which searches in the device list for a
> particular one by its vendor/device ID. This function is
> referenced in 127 files.
>
Again, in pci-driver.c:
static int pci_device_probe(struct device * dev)
{
int error = 0;
struct pci_driver *drv;
struct pci_dev *pci_dev;
drv = to_pci_driver(dev->driver);
pci_dev = to_pci_dev(dev);
pci_dev_get(pci_dev);
error = __pci_device_probe(drv, pci_dev);
if (error)
pci_dev_put(pci_dev);
return error;
}
Again, doing as Manohar suggested.
For those drivers using pci_get_device() - P313 of the Linux Device Drivers
3rd Ed, under "Old-style PCI Probing". Not a mechanism currently supported for
the VME bus.
> USB: usb_get_dev() referenced in 75 files.
>
I've already explained that I don't think the USB bus is good for comparison
due to very real differences in bus topology and use.
> There are also lots of direct calls to get_device() from .probe
> methods of devices not tied to a particular bus.
>
Which may therefore have nothing to do with a bus.
> Guess that was enough grepping.
>
It seems not quite.
>> Suitable bug fixes are welcome.
>
> I sent a fix (as part of an admittedly large patchset) in
> Nov 2010[1], ie 9 months ago, you were sick at the time and
> told me you'd have a look at the changes later[2], which
> unfortunately never happened, even after pinging you off-list.
>
Some of those patches were applied:
http://git.kernel.org/gitweb.cgi?p=linux/kernel/git/gregkh/staging-2.6.git;a=history;f=drivers/staging/vme;hb=HEAD
The remaining patches completely changed the VME bus model. I said at the time
I wasn't overly happy with the model you had put forward. Take this as my review:
- I am not happy with the proposed alternative model.
Sorry for not getting back to you sooner. I'm afraid my TODO list sometimes
ends up acting as a stack rather than a FIFO.
> Anyway let's forget that, Manohar's patches are what matters now.
>
Will do.
Martyn
--
Martyn Welch (Principal Software Engineer) | Registered in England and
GE Intelligent Platforms | Wales (3828642) at 100
T +44(0)127322748 | Barbirolli Square, Manchester,
E martyn.welch@xxxxxx | M2 3AB VAT:GB 927559189
_______________________________________________
devel mailing list
devel@xxxxxxxxxxxxxxxxxxxxxx
http://driverdev.linuxdriverproject.org/mailman/listinfo/devel
