On 11/4/2021 3:44 PM, Keith Busch wrote:
> External email: Use caution opening links or attachments
>
>
> On Wed, Nov 03, 2021 at 11:46:29PM -0700, Chaitanya Kulkarni wrote:
>> +static inline blk_status_t nvme_setup_verify(struct nvme_ns *ns,
>> + struct request *req, struct nvme_command *cmnd)
>> +{
>
> Due to recent driver changes, you need a "memset(cmnd, 0, sizeof(*cmnd))"
> prior to setting up the rest of the command, or you need to set each
> command dword individually.
Agree, will fix that in V1.
>
>> + cmnd->verify.opcode = nvme_cmd_verify;
>> + cmnd->verify.nsid = cpu_to_le32(ns->head->ns_id);
>> + cmnd->verify.slba =
>> + cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
>> + cmnd->verify.length =
>> + cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
>> + cmnd->verify.control = 0;
>> + return BLK_STS_OK;
>> +}
>
>> +static void nvme_config_verify(struct gendisk *disk, struct nvme_ns *ns)
>> +{
>> + u64 max_blocks;
>> +
>> + if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_VERIFY))
>> + return;
>> +
>> + if (ns->ctrl->max_hw_sectors == UINT_MAX)
>> + max_blocks = (u64)USHRT_MAX + 1;
>> + else
>> + max_blocks = ns->ctrl->max_hw_sectors + 1;
>
> If supported by the controller, this maximum is defined in the non-mdts
> command limits in NVM command set specific Identify Controller VSL field.
>
I need take a closer look at this. I'll fix that in V1.
>> +
>> + /* keep same as discard */
>> + if (blk_queue_flag_test_and_set(QUEUE_FLAG_VERIFY, disk->queue))
>> + return;
>> +
>> + blk_queue_max_verify_sectors(disk->queue,
>> + nvme_lba_to_sect(ns, max_blocks));
>> +
>> +}
Thanks for the comment Keith.