Re: [PATCH RFC] char: misc: Enforce simple minor space division

[Zijun Hu <zijun_hu@icloud.com>]

On 2025/6/24 23:50, Greg Kroah-Hartman wrote:
> On Fri, Jun 20, 2025 at 10:53:32PM +0800, Zijun Hu wrote:
>> From: Zijun Hu <zijun.hu@oss.qualcomm.com>
>>
>> Enforce simple minor space division related to macro MISC_DYNAMIC_MINOR
>> defined as 255 currently:
>>
>> <  255 : Fixed minor codes
>> == 255 : Indicator to request dynamic minor code
>>>  255 : Dynamic minor codes requested
> 
> Is this the rule today?  If so, does the now-added tests we have for
> misc device properly test for this?
> 

1) yes. this simple division becomes possible with recent commits below:
Commit: 31b636d2c416 ("char: misc: restrict the dynamic range to exclude
reserved minors")
Commit: c876be906ce7 ("char: misc: register chrdev region with all
possible minors")

both available fixed and dynamic minors interleaves with narrow space
[0, 255) before above commits.

it is easy to balance minor space division by adjusting macro
@MISC_DYNAMIC_MINOR if required in future as well.

Also hope all fixed minors are registered with header linux/miscdevice.h

2) no. below recent commit don't cover the simple division fully.
Commit: 74d8361be344 ("char: misc: add test cases")

drivers/misc/misc_minor_kunit.c may need to be corrected to reflecting
division today.

>> This enforcing division also solves misc_register() reentry issue below:
>>
>> // Suppose both static @dev_A and @dev_B want to request dynamic minors.
>> @dev_A.minor(255) @dev_B.minor(255)
>>
>> // Register @dev_A then de-register it.
>> @dev_A.minor(255) -> registered -> @dev_A.minor(500) -> de-registered
>> -> @dev_A.minor(500)
>>
>> // Register @dev_B
>> @dev_B.minor(255) -> registered -> @dev_B.minor(500)
>>
>> // Register @dev_A again
>> @dev_A.minor(500) -> encounter -EBUSY error since @dev_B has got 500.
> 
> Does this ever really happen?
> 

i never meet this issue. but in theory, it may happen as explained below:

misc_register()/misc_deregister() are sometimes called within driver's
probe()/remove(), such cases have reentry requirements

actually, error handling in misc_register() also reset minor code allocated:

	if (IS_ERR(misc->this_device)) {
		misc_minor_free(misc->minor);
		if (is_dynamic) {
			misc->minor = MISC_DYNAMIC_MINOR;
		}
		err = PTR_ERR(misc->this_device);
		goto out;
	}

> And with the recent changes in the last dev cycle in this code area, is
> it still an issue?
>

this is a different issue with the ones recent changes fix.

>> Side effects:
>> It will be refused to register device whose fixed minor > 255.
> 
> Do we have any in-kernel users that are > 255?

NO. no kernel users have such usage.

Actually, if fixed minor (>255) is used to register miscdev, it may
encounter failure since the fixed minor (>255) may be allocated for
other dynamic requests.

> 
> thanks,
> 
> greg k-h