Re: SCSI API generics (was Re: [PATCH] add block/unblock to iscsi class)

[Mike Christie <mikenc@us.ibm.com>]

Mike Christie wrote:
> Andrew Vasquez wrote:
> 
>> On Wed, 23 Feb 2005, Mike Christie wrote:
>>
>>
>>> Andrew Vasquez wrote:
>>>
>>>> Speaking of which, are there any major objections to the patches
>>>> proposed here:
>>>>
>>>> Add scsi_target_[un]block() and scsi_target_remove() generics
>>>> http://marc.theaimsgroup.com/?l=linux-scsi&m=110867050306738&w=2
>>>>
>>>> [RFC] adding per scsi-host workqueues for defered processing
>>>> http://marc.theaimsgroup.com/?l=linux-scsi&m=110903148407438&w=2
>>>>
>>>
>>> When using single_threaded_workqueues do all single threaded ones
>>> in the system use the same cpu_workqueue_struct? Will this be
>>> a potential bottleneck if we end up using the workqueue for more
>>> operations in the future?
>>
>>
>>
>> The kernel-thread that is created for a single-threaded workqueue is
>> not bound to any particular CPU:
>>
>>     kthread.c:kthread()
>>
>>     ...   
>>         /* By default we can run anywhere, unlike keventd. */
>>         set_cpus_allowed(current, CPU_MASK_ALL);
>>
> 
> My question is not about which cpu it is run on.
> I am asking about the cpu_workqueue_struct. In
> create_workqueue_thread cpu always equals zero so
> later when it is run it can be on any cpu but every
> single threaded workqueue will access the same
> cpu_workqueue_struct.
> 

Oh wait nevermind that too. I should read the code more
closely next time.