Re: blk_iopoll_enabled

[Sagi Grimberg <sagig@dev.mellanox.co.il>]

On 3/13/2014 5:34 PM, Jens Axboe wrote:
> On 03/12/2014 09:40 PM, Jens Axboe wrote:
>> On 2014-03-12 11:35, Sagi Grimberg wrote:
>>> Hey Jens,
>>>
>>> So I'm looking at blk_iopoll for iSER & SRP and I have one gap I would
>>> like to get your input on.
>>>
>>> In blk-iopoll.c I see:
>>> int blk_iopoll_enabled = 1;
>>> EXPORT_SYMBOL(blk_iopoll_enabled);
>>>
>>> This is set to 1 and exported for everyone to condition and *modify* 
>>> it.
>>> The documentation says the user must check this when using blk_iopoll.
>>>
>>> My question is why? it should always be true shouldn't it?
>>> Moreover, isn't it better to protect it using a get() function of it?
>>> can't someone set it to 0 and shutdown the lights on everyone else?
>>
>> We can kill it. IIRC, it was mostly used during development to test
>> functionality and performance (and wired to a sysfs toggle file).
>>
>> But... Just because it's exported doesn't mean that random users are
>> allowed to just disable it. What's left now is an always-on, so we can
>> remove it.
>
> It _is_ still exported, I just forgot it was put in sysctl. I'll kill 
> it off completely now. It's safe to toggle from user space, but only 
> practically so if IO is idle. It makes no attempt to ensure that, as 
> it was just a debug thing.
>

Thanks

Sagi.