Re: [RFC PATCH 3/3] hw/block/nvme: add nvme_inject_state HMP command

[Minwoo Im]

On 21-02-11 12:00:11, Keith Busch wrote:
> On Thu, Feb 11, 2021 at 04:52:52AM +0900, Minwoo Im wrote:
> > nvme_inject_state command is to give a controller state to be.
> > Human Monitor Interface(HMP) supports users to make controller to a
> > specified state of:
> > 
> >     normal:                 Normal state (no injection)
> >     cmd-interrupted:        Commands will be interrupted internally
> > 
> > This patch is just a start to give dynamic command from the HMP to the
> > QEMU NVMe device model.  If "cmd-interrupted" state is given, then the
> > controller will return all the CQ entries with Command Interrupts status
> > code.
> > 
> > Usage:
> >     -device nvme,id=nvme0,....
> > 
> >     (qemu) nvme_inject_state nvme0 cmd-interrupted
> > 
> >     <All the commands will be interrupted internally>
> > 
> >     (qemu) nvme_inject_state nvme0 normal
> > 
> > This feature is required to test Linux kernel NVMe driver for the
> > command retry feature.
> 
> Once the user sets the injected state, all commands return that status
> until the user injects the normal state, so the CRD time is meaningless
> here. If we're really going this route, the state needs to return to
> normal on it's own.

That would also be fine to me.

> But I would prefer to see advanced retry tied to real errors that can be
> retried, like if we got an EBUSY or EAGAIN errno or something like that.

I have seen a thread [1] about ACRE.  Forgive me If I misunderstood this
thread or missed something after this thread.  It looks like CRD field in
the CQE can be set for any NVMe error state which means it *may* depend on
the device status.  And this patch just introduced a internal temporarily
error state of the controller by returning Command Intrrupted status.

I think, in this stage, we can go with some errors in the middle of the
AIO (nvme_aio_err()) for advanced retry.  Shouldn't AIO errors are
retry-able and supposed to be retried ?

> The interface you found to implement this is very interesting though.

Thanks, I just wanted to suggest a scheme to inject something to a
running NVMe device model for various testing.

[1] https://www.spinics.net/lists/dm-devel/msg42165.html