[PATCH 1/2] nvme: only allow entering LIVE from CONNECTING state

[Daniel Wagner]

On Sun, Apr 27, 2025 at 08:59:13AM -0700, Guenter Roeck wrote:
> Hi,
> 
> On Fri, Feb 14, 2025 at 09:02:03AM +0100, Daniel Wagner wrote:
> > The fabric transports and also the PCI transport are not entering the
> > LIVE state from NEW or RESETTING. This makes the state machine more
> > restrictive and allows to catch not supported state transitions, e.g.
> > directly switching from RESETTING to LIVE.
> > 
> > Signed-off-by: Daniel Wagner <wagi at kernel.org>
> 
> nvme_handle_aen_notice(), when handling NVME_AER_NOTICE_FW_ACT_STARTING,
> sets the state to RESETTING and and triggers a worker. This worker
> waits for firmware activation to complete and then tries to set the
> state back to LIVE. This step now fails.
> 
> Possibly the handling of NVME_AER_NOTICE_FW_ACT_STARTING needs to be
> improved. However, leaving the NVME in RESETTING state after an
> NVME_AER_NOTICE_FW_ACT_STARTING event is worse.
> 
> I think this patch should be reverted at least for the time being until
> the handling of NVME_AER_NOTICE_FW_ACT_STARTING no longer relies on a
> direct state change from RESETTING to LIVE.

ee59e3820ca9 ("nvme-fc: do not ignore connectivity loss during connecting")
f13409bb3f91 ("nvme-fc: rely on state transitions to handle connectivity loss")

are depending on the fact that is not possible to switch from
NEW/RESETTING directly into LIVE.

I think it would be better to fix the worker instead dropping this patch
and the above fix for the fc transport.

What about:


diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index b502ac07483b..d3c4eacf607f 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -4493,7 +4493,8 @@ static void nvme_fw_act_work(struct work_struct *work)
                msleep(100);
        }

-       if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
+       if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_CONNECTING) ||
+           !nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
                return;

        nvme_unquiesce_io_queues(ctrl);