[PATCH V9 4/9] nvmet: add ZBD over ZNS backend support
Chaitanya Kulkarni
Chaitanya.Kulkarni at wdc.com
Tue Jan 12 23:57:15 EST 2021
On 1/11/21 23:48, Christoph Hellwig wrote:
>> diff --git a/drivers/nvme/target/admin-cmd.c b/drivers/nvme/target/admin-cmd.c
>> index a50b7bcac67a..bdf09d8faa48 100644
>> --- a/drivers/nvme/target/admin-cmd.c
>> +++ b/drivers/nvme/target/admin-cmd.c
>> @@ -191,6 +191,15 @@ static void nvmet_execute_get_log_cmd_effects_ns(struct nvmet_req *req)
>> log->iocs[nvme_cmd_dsm] = cpu_to_le32(1 << 0);
>> log->iocs[nvme_cmd_write_zeroes] = cpu_to_le32(1 << 0);
>> break;
>> + case NVME_CSI_ZNS:
>> + if (IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
>> + u32 *iocs = log->iocs;
>> +
>> + iocs[nvme_cmd_zone_append] = cpu_to_le32(1 << 0);
>> + iocs[nvme_cmd_zone_mgmt_send] = cpu_to_le32(1 << 0);
>> + iocs[nvme_cmd_zone_mgmt_recv] = cpu_to_le32(1 << 0);
>> + }
>> + break;
> We need to return errors if the command set is not actually supported.
> I also think splitting this into one helper per command set would
> be nice.
>
Okay.
>> @@ -644,6 +653,17 @@ static void nvmet_execuIt should be te_identify_desclist(struct nvmet_req *req)
>> if (status)
>> goto out;
>>
>> + if (IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
>> + u16 nvme_cis_zns = NVME_CSI_ZNS;
>> +
>> + if (req->ns->csi == NVME_CSI_ZNS)
>> + status = nvmet_copy_ns_identifier(req, NVME_NIDT_CSI,
>> + NVME_NIDT_CSI_LEN,
>> + &nvme_cis_zns, &off);
>> + if (status)
>> + goto out;
>> + }
> We need to add the CSI for every namespace, i.e. something like:
>
> status = nvmet_copy_ns_identifier(req, NVME_NIDT_CSI, NVME_NIDT_CSI_LEN,
> &req->ns->csi);
> if (status)
> goto out;
>
> and this hunk needs to go into the CSI patch.
even better, we can get rid of the local variables...
>> if (sg_zero_buffer(req->sg, req->sg_cnt, NVME_IDENTIFY_DATA_SIZE - off,
>> off) != NVME_IDENTIFY_DATA_SIZE - off)
>> status = NVME_SC_INTERNAL | NVME_SC_DNR;
>> @@ -660,8 +680,16 @@ static void nvmet_execute_identify(struct nvmet_req *req)
>> switch (req->cmd->identify.cns) {
>> case NVME_ID_CNS_NS:
>> return nvmet_execute_identify_ns(req);
>> + case NVME_ID_CNS_CS_NS:
>> + if (req->cmd->identify.csi == NVME_CSI_ZNS)
>> + return nvmet_execute_identify_cns_cs_ns(req);
>> + break;
>> case NVME_ID_CNS_CTRL:
>> return nvmet_execute_identify_ctrl(req);
>> + case NVME_ID_CNS_CS_CTRL:
>> + if (req->cmd->identify.csi == NVME_CSI_ZNS)
>> + return nvmet_execute_identify_cns_cs_ctrl(req);
>> + break;
> How does the CSI get mirrored into the cns field?
>
There is only one cns and one csi value we set from host/zns.c
This is just to reject req if we receive anything else or there is
any change on the host we fail.
>> diff --git a/drivers/nvme/target/core.c b/drivers/nvme/target/core.c
>> index 672e4009f8d6..17d5da062a5a 100644
>> --- a/drivers/nvme/target/core.c
>> +++ b/drivers/nvme/target/core.c
>> @@ -1107,6 +1107,7 @@ static inline u8 nvmet_cc_iocqes(u32 cc)
>> static inline bool nvmet_cc_css_check(u8 cc_css)
>> {
>> switch (cc_css <<= NVME_CC_CSS_SHIFT) {
>> + case NVME_CC_CSS_CSI:
>> case NVME_CC_CSS_NVM:
>> return true;
>> default:
>> @@ -1173,6 +1174,8 @@ static void nvmet_init_cap(struct nvmet_ctrl *ctrl)
>> {
>> /* command sets supported: NVMe command set: */
>> ctrl->cap = (1ULL << 37);
>> + if (IS_ENABLED(CONFIG_BLK_DEV_ZONED))
>> + ctrl->cap |= (1ULL << 43);
>> /* CC.EN timeout in 500msec units: */
>> ctrl->cap |= (15ULL << 24);
>> /* maximum queue entries supported: */
> This needs to go into a separate patch for multiple command set support.
> We can probably merge the CAP and CC bits with the CSI support, though.
Do you mean previous patch ? but we don't add handlers non-default I/O
command set until this patch..
>> + if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) && bdev_is_zoned(ns->bdev)) {
> bdev_is_zoned should be probably stubbed out for !CONFIG_BLK_DEV_ZONED
> these days.
Are you saying something like following in the prep patch ?or should
just remove
theIS_ENABLED(CONFIG_BLK_DEV_ZONED)part in above if?
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 028ccc9bdf8d..124086c1a0ba 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -1570,6 +1570,9 @@ static inline bool bdev_is_zoned(struct
block_device *bdev)
{
struct request_queue *q = bdev_get_queue(bdev);
+ if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED))
+ return false;
+
if (q)
return blk_queue_is_zoned(q);
>> +/*
>> + * ZNS related command implementation and helpers.
>> + */
> Well, that is the description of the whole file, isn't it? I don't think
> this comment adds much value.
Stupid comment, will remove it.
>> + /*
>> + * For ZBC and ZAC devices, writes into sequential zones must be aligned
>> + * to the device physical block size. So use this value as the logical
>> + * block size to avoid errors.
>> + */
> I do not understand the logic here, given that NVMe does not have
> conventional zones.
>
It should be :-
/*
* For ZBC and ZAC devices, writes into sequential zones must be aligned
* to the device physical block size. So use this value as the *physical*
* block size to avoid errors.
*/
More information about the Linux-nvme
mailing list