From mboxrd@z Thu Jan  1 00:00:00 1970
From: swise@opengridcomputing.com (Steve Wise)
Date: Tue, 29 May 2018 12:48:53 -0500
Subject: Question on inline support
In-Reply-To: <20180529174442.GA1637@lst.de>
References: <5276f947-0ccf-5d0e-53bc-724f65c1c837@opengridcomputing.com>
 <20180529174442.GA1637@lst.de>
Message-ID: <72e73b2e-4e3d-a39e-8514-a0a506a5b409@opengridcomputing.com>



On 5/29/2018 12:44 PM, hch@lst.de wrote:
> On Tue, May 29, 2018@12:33:56PM -0500, Steve Wise wrote:
>> Hey guys,
>>
>> While testing the nvme rdma target with inline_data_size=0 for my patch
>> series [1], I found that the nvme host side rejected the queue
>> connections with this error:
>>
>> "nvme nvme0: Mandatory keyed sgls are not support"
>>
>> Due to this code in nvme/host/rdma.c:
>>
>> ?????? /* sanity check keyed sgls */
>> ?????? if (!(ctrl->ctrl.sgls & (1 << 20))) {
>> ?????????????? dev_err(ctrl->ctrl.device, "Mandatory keyed sgls are not
>> support\n");
>> ?????????????? ret = -EINVAL;
>> ?????????????? goto out_remove_admin_queue;
>> ?????? }
> Keyed SGLs are bit 2, so the above should check for bit 2 as well.

Ah.? I'll fix that.

> We should still check for bit 20 if inline data is supported, though.
>

Why, exactly?? Use of inline data is determined by the target's
advertised ioccsz, and handled in nvme/host/rdma.c:

static inline size_t nvme_rdma_inline_data_size(struct nvme_rdma_queue
*queue)
{
??????? return queue->cmnd_capsule_len - sizeof(struct nvme_command);
}

Are you saying that the host should support a target that doesn't set
bit 20, yet advertises an ioccsz that could support inline??

Thanks,

Steve.