Commit a89fcca8 authored Aug 14, 2019 by Guilherme G. Piccoli Committed by Jens Axboe Aug 20, 2019

nvme: Fix cntlid validation when not using NVMEoF

Commit 1b1031ca ("nvme: validate cntlid during controller initialisation")
introduced a validation for controllers with duplicate cntlid that runs
on nvme_init_subsystem(). The problem is that the validation relies on
ctrl->cntlid, and this value is assigned (from id_ctrl value) after the
call for nvme_init_subsystem() in nvme_init_identify() for non-fabrics
scenario. That leads to ctrl->cntlid always being 0 in case we have a
physical set of controllers in the same subsystem.

This patch fixes that by loading the discovered cntlid id_ctrl value into
ctrl->cntlid before the subsystem initialization, only for the non-fabrics
case. The patch was tested with emulated nvme devices (qemu) having two
controllers in a single subsystem. Without the patch, we couldn't make
it work failing in the duplicate check; when running with the patch, we
could see the subsystem holding both controllers.

For the fabrics case we see ctrl->cntlid has a more intricate relation
with the admin connect, so we didn't change that.

Fixes: 1b1031ca ("nvme: validate cntlid during controller initialisation")
Signed-off-by: Guilherme G. Piccoli <gpiccoli@canonical.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

parent 504db087

drivers/nvme/host/core.c

+3 −1

Original line number	Diff line number	Diff line
		@@ -2597,6 +2597,9 @@ int nvme_init_identify(struct nvme_ctrl *ctrl)
		goto out_free;
		}

		if (!(ctrl->ops->flags & NVME_F_FABRICS))
		ctrl->cntlid = le16_to_cpu(id->cntlid);

		if (!ctrl->identified) {
		int i;

		@@ -2697,7 +2700,6 @@ int nvme_init_identify(struct nvme_ctrl *ctrl)
		goto out_free;
		}
		} else {
		ctrl->cntlid = le16_to_cpu(id->cntlid);
		ctrl->hmpre = le32_to_cpu(id->hmpre);
		ctrl->hmmin = le32_to_cpu(id->hmmin);
		ctrl->hmminds = le32_to_cpu(id->hmminds);

Admin message