Re: [PATCH 1/5] bfa: Brocade BFA FC SCSI driver (bfad)

[Mike Christie <michaelc@cs.wisc.edu>]

Jing Huang wrote:
>>> +
>>> +	im_port->shost->hostdata[0] = (unsigned long)im_port;
>>
>>
>> I think I asked about this before. Why do you just not allocate the
> bfad
>> or im_port in the hostdata? What is the struct hierarchy? Do you have
>> one bfad for the entire HBA/pci_device then have a im_port for each
> port
>> on the hba?
>>
> 
> Thanks for review the code again. Yes, we have one bfad for each HBA
> port/pci device, and one im_port (im stands for initiator mode) for each
> scsi_host/vport. We can certainly allocate im_port in scsi_host_alloc()
> as you suggest, but currently we allocate all possible fc4 port
> (initiator/tgt/ipfc) in one function, and it is just more convenient to
> do it in one place.

I think the refcounting is off though. For example in the shutdown case, 
if someone is accessing a scsi host sysfs attr, bfad_im_scsi_host_free 
would release the drivers refcounts on the scsi_host and free the 
im_port, but the sysfs code could still be accessing the im port. The 
shost would still be there because the sysfs code took a refcount on 
kobject/kref, but the im port is now gone. When you allocate the port 
struct in the hostdata with scsi_host_alloc this type of issue is 
handled for you.