Re: [PATCH linux-2.6.18-xen] blktap: make max # of tap devices a module parameter

["Jan Beulich" <JBeulich@novell.com>]

>>> On 22.02.11 at 18:44, Daniel Stodden <daniel.stodden@citrix.com> wrote:
> On Tue, 2011-02-22 at 12:34 -0500, Laszlo Ersek wrote:
>> On 02/22/11 16:49, Jan Beulich wrote:
>> >>>> On 22.02.11 at 15:20, Laszlo Ersek<lersek@redhat.com>  wrote:
>> >> Hi,
>> >>
>> >> should anybody still use the blktap(1) driver in linux-2.6.18-xen, the
>> >> following patch intends to make the maximum number of tapdevs
>> >> configurable at module insertion time. The number is clamped to [256 ..
>> >> NR_EVENT_CHANNELS]. I removed the definition of MAX_DEV_NAME because it
>> >> didn't seem to be used at all.
>> >>
>> >> Thanks for considering,
>> >> Laszlo Ersek
>> >
>> > Without replacing the call to register_chrdev() with one to
>> > __register_chrdev() (available only with 2.6.32 and newer) I
>> > can't see how you would get beyond 256 devices with the
>> > changes you propose.
>> 
>> Oops, sorry; I naively assumed that minor device numbers were already 
>> covered by an earlier change.
>> 
>> I figure register_chrdev() could be reimplemented in blktap, based on 
>> lower-level char_dev.c (and kobject) primitives, but I'm not sure if the 
>> original goal is worth that ugliness. In any case, should I bother 
>> posting a version like that eventually, or would it have no chance of 
>> being accepted?
> 
> I'm pretty sure minors > 256 date way before 2.6.32. Here's the module
> init fragment from blktap2, replacing the register_chrdev() call:

Sure, just that you have do more things "manually".

> int __init
> blktap_ring_init(void)
> {
> 	dev_t dev = 0;
> 	int err;
> 
> 	cdev_init(&blktap_ring_cdev, &blktap_ring_file_operations);
> 	blktap_ring_cdev.owner = THIS_MODULE;
> 
> 	err = alloc_chrdev_region(&dev, 0, MAX_BLKTAP_DEVICE, "blktap2");
> 	if (err < 0) {
> 		BTERR("error registering ring devices: %d\n", err);
> 		return err;
> 	}
> 
> 	err = cdev_add(&blktap_ring_cdev, dev, MAX_BLKTAP_DEVICE);
> 	if (err) {
> 		BTERR("error adding ring device: %d\n", err);
> 		unregister_chrdev_region(dev, MAX_BLKTAP_DEVICE);
> 		return err;
> 	}
> 
> 	blktap_ring_major = MAJOR(dev);
> 	BTINFO("blktap ring major: %d\n", blktap_ring_major);
> 
> 	return 0;
> }

Any reason why in .32 and newer you still don't use
__register_chrdev() (and __unregister_chrdev())? And even
if this still needs to be this way, I note that unregister_chrdev()
calls __unregister_chrdev_region() before cdev_del(), while
blktap_ring_exit() does it the other way around?

Jan