Re: [PATCH 06/12] block: lift setting the readahead size into the block layer

On Thu, Sep 17, 2020 at 12:35:40PM +0200, Jan Kara wrote:

quoted

diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
index 81722cdcf0cb21..95eb35324e1a61 100644
--- a/block/blk-sysfs.c
+++ b/block/blk-sysfs.c

@@ -245,7 +245,6 @@ queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
 
 	spin_lock_irq(&q->queue_lock);
 	q->limits.max_sectors = max_sectors_kb << 1;
-	q->backing_dev_info->io_pages = max_sectors_kb >> (PAGE_SHIFT - 10);
 	spin_unlock_irq(&q->queue_lock);

So do I get it right that readahead won't now be limited if you store lower
value to max_sectors? Why? I'd consider io_pages a "cached value" of
max_sectors and thus expect it to change together with max_sectors...

Most to start untangling the bdi from the queue.  But I had to peddle
back on that in the follow on series anyway, so I can add this back.

quoted

@@ -812,7 +813,7 @@ static void __device_add_disk(struct device *parent, struct gendisk *disk,
 		disk->flags |= GENHD_FL_SUPPRESS_PARTITION_INFO;
 		disk->flags |= GENHD_FL_NO_PART_SCAN;
 	} else {
-		struct backing_dev_info *bdi = disk->queue->backing_dev_info;
+		struct backing_dev_info *bdi = q->backing_dev_info;
 		struct device *dev = disk_to_dev(disk);
 		int ret;

Not sure how/why these changes got here... Not that I care too much :)

Because more changes in this area in earlier versions of the patches.
But yes, this shouldn't be here, so I'll drop it.

quoted

@@ -407,7 +406,6 @@ aoeblk_gdalloc(void *vp)
 	WARN_ON(d->gd);
 	WARN_ON(d->flags & DEVFL_UP);
 	blk_queue_max_hw_sectors(q, BLK_DEF_MAX_SECTORS);
-	q->backing_dev_info->ra_pages = READ_AHEAD / PAGE_SIZE;
 	d->bufpool = mp;
 	d->blkq = gd->queue = q;
 	q->queuedata = d;

Shouldn't AOE set 2MB optimal IO size so that readahead is equivalent to
previous behavior?

Sure, I'll add a separate patch just for that.

quoted

diff --git a/drivers/md/bcache/super.c b/drivers/md/bcache/super.c
index 1bbdc410ee3c51..ff2101d56cd7f1 100644
--- a/drivers/md/bcache/super.c
+++ b/drivers/md/bcache/super.c

@@ -1427,10 +1427,6 @@ static int cached_dev_init(struct cached_dev *dc, unsigned int block_size)
 	if (ret)
 		return ret;
 
-	dc->disk.disk->queue->backing_dev_info->ra_pages =
-		max(dc->disk.disk->queue->backing_dev_info->ra_pages,
-		    q->backing_dev_info->ra_pages);
-

So bcache is basically stacking readahead here on top of underlying cache
device. I don't see this being replicated by your patch so it is lost now?
Probably this should be replaced by properly inheriting optimal IO size?

Yes, I'll add another patch.