Re: [PATCH v9 3/4] scsi: ufs: L2P map management for HPB read

[Bart Van Assche]

On 2020-08-28 00:19, Daejun Park wrote:
> +static unsigned int ufshpb_host_map_kbytes = 1024;

A comment that explains where this value comes from would be welcome.

> +static struct ufshpb_req *ufshpb_get_map_req(struct ufshpb_lu *hpb,
> +					     struct ufshpb_subregion *srgn)
> +{
> +	struct ufshpb_req *map_req;
> +	struct request *req;
> +	struct bio *bio;
> +
> +	map_req = kmem_cache_alloc(hpb->map_req_cache, GFP_KERNEL);
> +	if (!map_req)
> +		return NULL;
> +
> +	req = blk_get_request(hpb->sdev_ufs_lu->request_queue,
> +			      REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);

Why BLK_MQ_REQ_PREEMPT? Since this code is only executed while medium access
commands are processed and since none of these commands have the PREEMPT flag
set I think that the PREEMPT flag should be left out. Otherwise there probably
will be weird interactions with runtime suspended devices.

Is it acceptable that the above blk_get_request() call blocks if a UFS device
has been runtime suspended? If not, consider using the flag BLK_MQ_REQ_NOWAIT
instead.

> +	bio = bio_alloc(GFP_KERNEL, hpb->pages_per_srgn);
> +	if (!bio) {
> +		blk_put_request(req);
> +		goto free_map_req;
> +	}

If the blk_get_request() would be modified such that it doesn't wait, this
call may have to be modified too (GFP_NOWAIT?).

> +	if (rgn->rgn_state == HPB_RGN_INACTIVE) {
> +		if (atomic_read(&lru_info->active_cnt)
> +		    == lru_info->max_lru_active_cnt) {

When splitting a line, please put comparison operators at the end of the line
instead of at the start, e.g. as follows:

		if (atomic_read(&lru_info->active_cnt) ==
		    lru_info->max_lru_active_cnt) {

> +	pool_size = DIV_ROUND_UP(ufshpb_host_map_kbytes * 1024, PAGE_SIZE);

Please use PAGE_ALIGN() to align to the next page boundary.

Thanks,

Bart.