Re: [PATCH v3] cxlflash: Base support for IBM CXL Flash Adapter

[Manoj Kumar <manoj@linux.vnet.ibm.com>]

Brian:

Thanks for your review. Responses are inline below.

- Manoj Kumar


On 6/4/2015 9:38 AM, Brian King wrote:
>> +
>> +	write_lock(&cfg->tmf_lock);
>
> What is this lock protecting? The only thing it seems to be accomplishing is
> making sure one thread isn't sending a TMF and another thread is sending
> a normal I/O command at the exact same time, yet it looks like you still
> allow a TMF to be sent and a normal I/O to be sent immediately after, before
> receiving the TMF response.
>

Originally this section was waiting for the TMF response. I see that is 
no longer the case. Will restore the original behavior, with adequate 
locking.


>> +		afu->room = readq_be(&afu->host_map->cmd_room);
>
> Looks like you now have an MMIO load as part of sending every command,
> including commands coming from queuecommand. Won't that be a performance issue?
> Is there any way to avoid this? Could you perhaps decrement afu->room
> in this function and only re-read it from the AFU when the counter hits zero?
>

Good point. Will revise to avoid MMIO in this performance path.