Re: [PATCH net-next v4] ptp: ocp: Expose various resources on the timecard.

On Thu, Aug 05, 2021 at 06:03:26AM -0700, Jakub Kicinski wrote:

On Wed, 4 Aug 2021 16:52:23 -0700 Jonathan Lemon wrote:

quoted

quoted

quoted

+static int
+ptp_ocp_devlink_info_get(struct devlink *devlink, struct devlink_info_req *req,
+			 struct netlink_ext_ack *extack)
+{
+	struct ptp_ocp *bp = devlink_priv(devlink);
+	char buf[32];
+	int err;
+
+	err = devlink_info_driver_name_put(req, KBUILD_MODNAME);
+	if (err)
+		return err;
+
+	if (bp->pending_image) {
+		err = devlink_info_version_stored_put(req,
+						      "timecard", "pending");  

"pending" is not a version. It seems you're talking to the flash
directly, why not read the version?  

We're not talking to the flash yet.  We're writing a new image, but don't
know the image version, since it's not accessible from the FPGA blob.  So
since we're don't know what the stored image is until we reboot, I've set
it to 'pending' here - aka "pending reboot".  Could also be "unknown".

Having the driver remember that the device was flashed is not a solid
indication that the image is actually different. It may be that user
flashed the same version, driver may get reloaded and lose the
indication.. Let's not make a precedent for (ab) use of the version
field to indicate reset required.

I'd like to have some way to remind/tell the user that a reset is required.

Right now, I can only get the running version from the FPGA register, so
after flashing, there's no way for me to know what's on the flash (or if 
the flash write failed).  Setting "pending" or "reboot" works for most
cases - but obviously fails if the driver is reloaded. 

But most users won't do rmmod/insmod, just a reboot.

quoted

quoted

quoted

+	}
+
+	if (bp->image) {
+		u32 ver = ioread32(&bp->image->version);
+
+		if (ver & 0xffff) {
+			sprintf(buf, "%d", ver);
+			err = devlink_info_version_running_put(req,
+							       "timecard",
+							       buf);
+		} else {
+			sprintf(buf, "%d", ver >> 16);
+			err = devlink_info_version_running_put(req,
+							       "golden flash",
+							       buf);  

What's the difference between "timecard" and "golden flash"?  

There are two images stored in flash: "golden image", an image that
just provides flash functionality, and the actual featured FPGA image.

quoted

Why call firmware for a timecard timecard? We don't call NIC
firmware "NIC".  

I didn't see a standard string to use.  I can call it 'fw.version', just
needed to differentiate it between the 'golden flash' loader and actual
firmware.

Is the 'golden flash' a backup in case full featured image does not
work or 'first stage' image/'loader'? IIUC it's the latter so maybe
we can just use "loader"? "fw" for the actual image would be better 
than "fw.version" the entire string is a version after all.

Undetermined.  I believe that /currently/ it is the same image, but
that is just happenstance.  The vendor says the only functionality 
that the golden image guarantees is flashing - the rest of the FPGA
may or may not work.

I'll use "fw" and "loader".

quoted

quoted

quoted

+static void
+ptp_ocp_devlink_health_register(struct devlink *devlink)
+{
+	struct ptp_ocp *bp = devlink_priv(devlink);
+	struct devlink_health_reporter *r;
+
+	r = devlink_health_reporter_create(devlink, &ptp_ocp_health_ops, 0, bp);
+	if (IS_ERR(r))
+		dev_err(&bp->pdev->dev, "Failed to create reporter, err %ld\n",
+			PTR_ERR(r));
+	bp->health = r;
+}  

What made you use devlink health here? Why not just print that "No GPS
signal" message to the logs? Devlink health is supposed to give us
meaningful context dumps and remediation, here neither is used.  

The initial idea was to use 'devlink monitor' report the immediate
failure of the GNSS signal (rather than going through the kernel logs)
The 'devlink health' also keeps a count of how often the GPS signal
is lost.

Our application guys decided to use a different monitoring method,
so I can rip this out if objectionable. 

Great, thanks!