Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <4746a000-2220-211e-1bd6-79c15c18a85c@samsung.com>
Date:   Fri, 17 Jun 2022 07:45:57 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
        Thunderbird/91.9.1
Subject: Re: [PATCH v7 13/13] dm: add non power of 2 zoned target
Content-Language: en-US
To:     Damien Le Moal <damien.lemoal@opensource.wdc.com>, <hch@lst.de>,
        <snitzer@redhat.com>, <axboe@kernel.dk>
CC:     <bvanassche@acm.org>, <linux-kernel@vger.kernel.org>,
        <jiangbo.365@bytedance.com>, <hare@suse.de>, <pankydev8@gmail.com>,
        <dm-devel@redhat.com>, <jonathan.derrick@linux.dev>,
        <gost.dev@samsung.com>, <dsterba@suse.com>, <jaegeuk@kernel.org>,
        <linux-nvme@lists.infradead.org>, <Johannes.Thumshirn@wdc.com>,
        <linux-block@vger.kernel.org>,
        Damien Le Moal <damien.lemoal@wdc.com>
From:   Pankaj Raghav <p.raghav@samsung.com>
In-Reply-To: <0c4f30f2-c206-0201-31e3-fbb9edbdf666@opensource.wdc.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 7bit
CMS-TYPE: 201P
References: <20220615101920.329421-1-p.raghav@samsung.com>
        <CGME20220615102011eucas1p220368db4a186181b1927dea50a79e5d4@eucas1p2.samsung.com>
        <20220615101920.329421-14-p.raghav@samsung.com>
        <63b0cfb6-eb24-f058-e502-2637039c5a98@opensource.wdc.com>
        <0b819562-8b16-37b6-9220-28bf1960bccb@samsung.com>
        <0c4f30f2-c206-0201-31e3-fbb9edbdf666@opensource.wdc.com>
Precedence: bulk

On 2022-06-17 01:49, Damien Le Moal wrote:
>>> Why do you need to change dm_set_zones_restrictions() at all ?
>>>
>> When the device mapper is created, the q->limits gets inherited from the
>> underlying device. The chunk sectors of the target and the device will
>> be the same but we want the chunk sector of the target to be different
>> (rounded to po2) compared to the underlying device's chunk sector. This
>> needs to be done only for the dm-po2z target and not for other targets
>> that uses npo2 zoned devices (like dm-linear). So to perform this
>> operation in a target independent way in dm-zone.c, I chose to always
>> revalidate npo2 zoned device and update the chunk sector and nr_zones in
>> dm_zone_revalidate_cb based on the zone information from the target.
>> This allows to set the limits correctly for dm-po2z target.
> 
> But DM revalidate will be called for the target AFTER it is setup (after
> its gendisk is added). So how can DM revalidate see the incorrect zone
> size ? If that is the case, then the target constructor is broken or
> missing something. DM revalidate zone is generic and only allocates the
> zone bitmaps for the target device. There should be not need at all to
> touch that function.
> 
I think this is a cleaner approach using features flag and io_hints
instead of messing with the revalidate zone function:

diff --git a/drivers/md/dm-table.c b/drivers/md/dm-table.c
index 135c0cc190fb..c97a71e0473f 100644
--- a/drivers/md/dm-table.c
+++ b/drivers/md/dm-table.c
@@ -1618,6 +1618,9 @@ static int device_not_matches_zone_sectors(struct
dm_target *ti, struct dm_dev *
 	if (!blk_queue_is_zoned(q))
 		return 0;

+	if(dm_target_supports_emulated_zone_size(ti->type))
+		return 0;
+
 	return blk_queue_zone_sectors(q) != *zone_sectors;
 }

diff --git a/drivers/md/dm-zoned-npo2-target.c
b/drivers/md/dm-zoned-npo2-target.c
index dad135964e09..b203be808f09 100644
--- a/drivers/md/dm-zoned-npo2-target.c
+++ b/drivers/md/dm-zoned-npo2-target.c
@@ -187,6 +187,12 @@ static int dmz_npo2_end_io(struct dm_target *ti,
struct bio *bio,
 	return DM_ENDIO_DONE;
 }

+static void dmz_npo2_io_hints(struct dm_target *ti, struct queue_limits
*limits)
+{
+	struct dmz_npo2_target *dmh = ti->private;
+	limits->chunk_sectors = dmh->zsze_po2;
+}
+
 static int dmz_npo2_map(struct dm_target *ti, struct bio *bio)
 {
 	struct dmz_npo2_target *dmh = ti->private;
@@ -233,12 +239,13 @@ static int dmz_npo2_iterate_devices(struct
dm_target *ti,
 static struct target_type dmz_npo2_target = {
 	.name = "zoned-npo2",
 	.version = { 1, 0, 0 },
-	.features = DM_TARGET_ZONED_HM,
+	.features = DM_TARGET_ZONED_HM | DM_TARGET_EMULATED_ZONE_SIZE,
 	.map = dmz_npo2_map,
 	.end_io = dmz_npo2_end_io,
 	.report_zones = dmz_npo2_report_zones,
 	.iterate_devices = dmz_npo2_iterate_devices,
 	.module = THIS_MODULE,
+	.io_hints = dmz_npo2_io_hints,
 	.ctr = dmz_npo2_ctr,
 };

diff --git a/include/linux/device-mapper.h b/include/linux/device-mapper.h
index c2a3758c4aaa..9f3a4d98a22a 100644
--- a/include/linux/device-mapper.h
+++ b/include/linux/device-mapper.h
@@ -283,6 +283,15 @@ struct target_type {
 #define dm_target_supports_mixed_zoned_model(type) (false)
 #endif

+#ifdef CONFIG_BLK_DEV_ZONED
+#define DM_TARGET_EMULATED_ZONE_SIZE	0x00000400
+#define dm_target_supports_emulated_zone_size(type) \
+	((type)->features & DM_TARGET_EMULATED_ZONE_SIZE)
+#else
+#define DM_TARGET_EMULATED_ZONE_SIZE	0x00000000
+#define dm_target_supports_emulated_zone_size(type) (false)
+#endif
+
 struct dm_target {
 	struct dm_table *table;
 	struct target_type *type;