On Mon, Feb 08, 2021 at 03:01:28PM -0800, Sagi Grimberg wrote:

quoted

quoted

quoted

Hi Sagi,

Yes. The Discovery Controller currently allows it. The problem, however,
is that the DC seems to be expecting these values to be non-zero.

I tried setting IOCQES=0 and IOSQES=0, which the DC allows (i.e. Prop
Set return status=0). However, when I follow this by a "Property Get -
Controller Status", the Status now shows "Controller Fatal Status (CFS)"
(see Base specs - Figure 79).

Yes, the oversight extends to the target that expects it (shared code
with I/O controllers).

Does this fix your issue?

Seems like the Identify Controller SQES and CQES fields ought to be
updated too, but the spec isn't clear on this. Feels like ECN
material...

Yes, we should start with this patch though right?

Yep, your patch looks good.

I was just thinking a generic implementation could align the CC settings with
the ID_CTRL values rather than checking for a discovery controller, but I
noticed the spec didn't provide us that opportunity. That was actually a
misguided thought anyway since we don't even have ID_CTRL data on the initial
CC setting! :)

_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme