Re: [PATCH 1/4] block: add zone open, close and finish support

["Matias Bjørling" <mb@lightnvm.io>]

On 6/25/19 12:27 AM, Chaitanya Kulkarni wrote:
> On 6/24/19 12:43 PM, Bart Van Assche wrote:
>> On 6/21/19 6:07 AM, Matias Bjørling wrote:
>>> diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
>>> index 95202f80676c..067ef9242275 100644
>>> --- a/include/linux/blk_types.h
>>> +++ b/include/linux/blk_types.h
>>> @@ -284,13 +284,20 @@ enum req_opf {
>>>     	REQ_OP_DISCARD		= 3,
>>>     	/* securely erase sectors */
>>>     	REQ_OP_SECURE_ERASE	= 5,
>>> -	/* reset a zone write pointer */
>>> -	REQ_OP_ZONE_RESET	= 6,
>>>     	/* write the same sector many times */
>>>     	REQ_OP_WRITE_SAME	= 7,
>>>     	/* write the zero filled sector many times */
>>>     	REQ_OP_WRITE_ZEROES	= 9,
>>>     
>>> +	/* reset a zone write pointer */
>>> +	REQ_OP_ZONE_RESET	= 16,
>>> +	/* Open zone(s) */
>>> +	REQ_OP_ZONE_OPEN	= 17,
>>> +	/* Close zone(s) */
>>> +	REQ_OP_ZONE_CLOSE	= 18,
>>> +	/* Finish zone(s) */
>>> +	REQ_OP_ZONE_FINISH	= 19,
>>> +
>>>     	/* SCSI passthrough using struct scsi_request */
>>>     	REQ_OP_SCSI_IN		= 32,
>>>     	REQ_OP_SCSI_OUT		= 33,
>>> @@ -375,6 +382,22 @@ static inline void bio_set_op_attrs(struct bio *bio, unsigned op,
>>>     	bio->bi_opf = op | op_flags;
>>>     }
>>
>> Are the new operation types ever passed to op_is_write()? The definition
>> of that function is as follows:
>>
> May be we should change numbering since blktrace also relies on the
> having on_is_write() without looking at the rq_ops().
> 
> 197  * Data direction bit lookup
>    198  */
>    199 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
>    200                                  BLK_TC_ACT(BLK_TC_WRITE) };  <----
>    201
>    202 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
>    203 #define BLK_TC_PREFLUSH         BLK_TC_FLUSH
>    204
>    205 /* The ilog2() calls fall out because they're constant */
>    206 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
>    207           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ##
> __name))
>    208
>    209 /*
>    210  * The worker for the various blk_add_trace*() types. Fills out a
>    211  * blk_io_trace structure and places it in a per-cpu subbuffer.
>    212  */
>    213 static void __blk_add_trace(struct blk_trace *bt, sector_t sector,
> int bytes,
>    214                      int op, int op_flags, u32 what, int error,
> int pdu_len,
>    215                      void *pdu_data, union kernfs_node_id *cgid)
>    216 {
>    217         struct task_struct *tsk = current;
>    218         struct ring_buffer_event *event = NULL;
>    219         struct ring_buffer *buffer = NULL;
>    220         struct blk_io_trace *t;
>    221         unsigned long flags = 0;
>    222         unsigned long *sequence;
>    223         pid_t pid;
>    224         int cpu, pc = 0;
>    225         bool blk_tracer = blk_tracer_enabled;
>    226         ssize_t cgid_len = cgid ? sizeof(*cgid) : 0;
>    227
>    228         if (unlikely(bt->trace_state != Blktrace_running &&
> !blk_tracer))
>    229                 return;
>    230
>    231         what |= ddir_act[op_is_write(op) ? WRITE : READ];  <---
> 
> 
>> static inline bool op_is_write(unsigned int op)
>> {
>> 	return (op & 1);
>> }
>>
> 

The zone mgmt commands are neither write nor reads commands. I guess, 
one could characterize them as write commands, but they don't write any 
data, they update a state of a zone on a drive. One should keep it as 
is? and make sure the zone mgmt commands don't get categorized as either 
read/write.