Re: [PATCH 1/4] block: centrelize PI remapping logic to the block layer

[Max Gurtovoy <maxg@mellanox.com>]

On 9/4/2019 8:49 AM, Christoph Hellwig wrote:
> On Tue, Sep 03, 2019 at 01:21:59PM -0600, Jens Axboe wrote:
>> On 9/3/19 1:11 PM, Sagi Grimberg wrote:
>>>> +	if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
>>>> +	    error == BLK_STS_OK)
>>>> +		t10_pi_complete(req,
>>>> +				nr_bytes / queue_logical_block_size(req->q));
>>>> +
>>> div in this path? better to use  >> ilog2(block_size).
>>>
>>> Also, would be better to have a wrapper in place like:
>>>
>>> static inline unsigned short blk_integrity_interval(struct request *rq)
>>> {
>>> 	return queue_logical_block_size(rq->q);
>>> }
>> If it's a hot path thing that matters, I'd strongly suggest to add
>> a queue block size shift instead.
> Make that a protection_interval_shift, please.  While that currently
> is the same as the logical block size the concepts are a little
> different, and that makes it clear.  Except for that this patch looks
> very nice to me, it is great to avoid having drivers to deal with the
> PI remapping.

Christoph,

I was thinking about the following addition to the code (combination of 
all the suggestions):


diff --git a/block/blk-core.c b/block/blk-core.c
index 58ecfd3..cef604c 100644
--- a/block/blk-core.c
+++ b/block/blk-core.c
@@ -1409,7 +1409,7 @@ bool blk_update_request(struct request *req, 
blk_status_t error,
         if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
             error == BLK_STS_OK)
                 t10_pi_complete(req,
-                               nr_bytes / 
queue_logical_block_size(req->q));
+                               nr_bytes >> 
queue_logical_block_shift(req->q));

         if (unlikely(error && !blk_rq_is_passthrough(req) &&
                      !(req->rq_flags & RQF_QUIET)))
diff --git a/block/blk-settings.c b/block/blk-settings.c
index 2c18312..8183ffc 100644
--- a/block/blk-settings.c
+++ b/block/blk-settings.c
@@ -330,6 +330,7 @@ void blk_queue_max_segment_size(struct request_queue 
*q, unsigned int max_size)
  void blk_queue_logical_block_size(struct request_queue *q, unsigned 
short size)
  {
         q->limits.logical_block_size = size;
+       q->limits.logical_block_shift = ilog2(size);

         if (q->limits.physical_block_size < size)
                 q->limits.physical_block_size = size;
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 1ef375d..4a0115e 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -332,6 +332,7 @@ struct queue_limits {
         unsigned int            discard_alignment;

         unsigned short          logical_block_size;
+       unsigned short          logical_block_shift;
         unsigned short          max_segments;
         unsigned short          max_integrity_segments;
         unsigned short          max_discard_segments;
@@ -1267,6 +1268,16 @@ static inline unsigned int 
queue_max_segment_size(struct request_queue *q)
         return q->limits.max_segment_size;
  }

+static inline unsigned short queue_logical_block_shift(struct 
request_queue *q)
+{
+       unsigned short retval = 9;
+
+       if (q && q->limits.logical_block_shift)
+               retval = q->limits.logical_block_shift;
+
+       return retval;
+}
+
  static inline unsigned short queue_logical_block_size(struct 
request_queue *q)



_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme