Re: lock_kernel check...

[Dave Hansen <haveblue@us.ibm.com>]

Zwane Mwaikambo wrote:
> On Tue, 9 Jul 2002, Dave Hansen wrote:
> 
> 
>>It isn't absoulutely a bad thing to return while you have a lock held. 
>>    It might be hard to understand, or even crazy, but not immediately 
>>wrong :)
>>
>>// BKL protects both "a", and io port 0xF00D
>>bar()
>>{
>>	lock_kernel();
>>	return inb(0xF00D);
>>}
>>
>>int a;
>>foo()
>>{
>>	a = bar();
>>	a--;
>>	unlock_kernel();
>>}
> 
> But broken nonetheless, that kinda thing just looks ugly. Especially when 
> someone tries to call bar multiple times or consecutively or with the lock 
> already held or...

Yes, it is horribly ugly, but it is not broken.  As a function, if you 
document what you require your caller to do, there shouldn't be a 
problem.

Also, it is valid to have nested holds of the BKL.  You can never 
deadlock with another lock_kernel() which was done in the same process.

-- 
Dave Hansen
haveblue@us.ibm.com