[Nut-upsdev] [EXTERNAL] Re: Fixing Drops With SMART1500LCDXL & USB-HID Driver

[Charles Lepple]

On Jun 18, 2019, at 1:47 PM, David Zomaya wrote:
> 
> Charles and Wolfy,
>  
> Thanks for the replies.
>  
> I added some responses below.
> I think I got the driver debugging right, but let me know if it is off.
>  
> I should note:
> The CentOS 7.6 machine I have been testing with is a virtual machine (running on VMware ESXi 7.6). At least 1 customer and I have seen the same issue on VMs and physical boxes, so I don’t think that matters, but if it does let me know.
>  
>> “This is a little off-topic, but I would like to point out that not including a machine-readable serial number in the USB device descriptor makes it difficult for people to reliably use two or more UPSes on a single *nix system. Due to some complications with the way that USB devices are opened in libusb, there is no easy way to "open the next unused USB device", so we recommend that people match against the serial number.”
>>  
> Noted. I’ll kick this around internally and follow up. I know other units we make report the serial number. Maybe it’s a 2012 thing.

I think we saw a similar problem with APC UPSes not having string descriptors when attached to a VM, so that is worth checking on a physical box. If so, I apologize for jumping to the conclusion that the UPS was not providing the serial number.

>   
> “It would be interesting to see the debug log from usbhid-ups as well. It would give a little more context to the kernel errors. I haven't used a physical CentOS or RedHat system in a while, so I am not sure of the specifics needed to just stop the usbhid-ups driver, but then you can restart it with a few "-D" flags (3 should be sufficient for this kind of problem) and "-a TrippLiteUPS" to match this configuration. Please compress any log files (gzip preferred; zip works).”
>  
> Attached. This is with the settings and udev rules mentioned in the earlier. I can remove and redo if useful.

The log format looks okay, but I may not have been clear that I was looking for the log at the same time as when the kernel errors occur. So without the udev rules would be good.
 
> 
> “pollinterval defaults to 2, and to be honest, for most other UPSes, we suggest that people raise the value (since many UPSes do not update their filtered values more frequently than that anyway).”
>  
> Good to know on the default “2”. Reading upsmon.conf, I thought it was related to “POLLFREQ” which defaults to 30.

We've reworked some of the documentation for the next release, but the short answer is that USB HID drivers poll the essential status bits every `pollinterval` seconds ("Quick update..." in the log), and then grab the rest every `pollfreq` seconds ("Full update").

>  
>  “Do you know how frequently the Windows software polls the UPS?” 
>  
> For our PowerAlert Local software, generally we do half a second for USB (I’d need to check if that holds for all protocols, but I cannot think of any exceptions off the top of my head). In this particular case we’ve just been testing with baked-in Windows Power Options as a reference point here, so I’ll need to look into how frequently Windows polls.
 
Thanks, that is useful to know.

> “Should this be applied to other models as well, or just protocol 2012?”
>  
> I’ll dig into that and confirm. I’m not entirely convinced it is a protocol issue, but it very well could be.
 
(By "protocol" I am just referring to the idProduct field, rather than the protocol itself, since that is what the driver would eventually use to enable any special cases.)

>  
> “The 62-nut-usbups.rules file looks pretty standard. Do you know if the changes to 42-usb-hd-pm.rules are needed? It seems like none of the USB devices would have the right permissions if 62-nut-usbups.rules isn't sufficient (though this happened in Debian once).”
>  
> My means of testing wasn’t the most rigorous, but I did try to use variable isolation with these changes and some other changes. I could not make the drops stop without having all 3 of these changes present. I believe a web search lead me to this udev rule so I’ll dig up the link for context.

This is starting to make sense, though. The link would be helpful, but no worries if you can't find it.

I think you mentioned the CentOS version - which kernel version does that run? ("uname -r" is probably sufficient)

I will try to read up on the USB power management settings that the udev file is changing.

 
> “Note that NUT 2.7.4 has been out for some time now.”
>  
> Wolfy nailed this one, I just installed whatever the repository gave me. Should I test with 2.7.4?
 
Might not be necessary.

>> “I thought we did, but maybe I am confusing it with protocol 3016 devices. We actually added a lot of the protocol 2012 devices to the hardware compatibility list based on the test results that Eric provided, so I assume they worked then (about six years ago).
>>  
>> The protocol 3016 devices (in particular, the SMART1500LCDT and OMNI1500LCDT) sometimes don't even stay on USB long enough to read a USB descriptor, and this does seem correlated with newer motherboards. Example: https://github.com/networkupstools/nut/issues/577
>>  
>> From where I stand, there really shouldn't be anything that a user-space program (like a NUT driver) can do that should be able to cause a USB device to disconnect during normal polling. (Aside from firmware updates, which we don't attempt.) That said, I recognize that USB Phy layer problems can be hard to diagnose, and power management can compound the issue.”
>>  
> Interesting threads. Are you still seeing issues with LCDTs or have they subsided?

I moved my SMART1500LCDT off of the always-on system, so I don't have continuous data for it. I don't think it worked on Raspbian stretch the last time I tried. Others have rigged up scripts to reset a USB hub to simulate re-plugging the UPS USB connection.
>  
> “Some users prefer not to post the entire serial numbers from their UPS when reporting issues. Is there a convention for the serial number digits such that we can ask for just the first few digits, and get an idea as to whether the problem is limited to a given hardware or firmware revision? There seems to be a firmware revision buried in the HID descriptor for some models, but I don't know how to interpret it, and some of these connection problems present themselves before the UPS can return that HID report.”
>  
> Our serial numbers break down like this:
> https://www.tripplite.com/support/identify-products
> If they give you the first 13 characters of the serial number, you’d know the SKU and the datecode without having the full serial number. Firmware isn’t inherently baked into that though. i.e. you could have the same firmware on different SKUs.
> Does this help at all?
 
Very useful, thanks!

> “Although these models are not as common, we still hear from users with non-HID-PDC-based USB devices (and some serial UPSes as well). Publicly-available protocol documents would help us write better drivers for those devices. If not, a better way to identify models with proprietary protocols would be useful.” 
>  
> Does publicly available protocol = needs to be accessible to anyone
> Or
> We provide you with protocol docs if you agree not to share?
> I can look into the latter. I feel like we should be able to help here.

I think that a lot of the content of the protocol docs would end up in the logic of the source code. Trying to write open-source code based on a document that can't be shared seems risky to me.

>> “Another thing is considering how users get started with NUT. Sometimes a user inherits an UPS on a given system, and they want to set up NUT to monitor it. Ideally, we'd like to have a way for them to quickly triage whether a particular UPS model will work, and before they have NUT installed, they will likely have "lsusb" or similar tools to enumerate devices. Other times, a user is replacing another UPS, and they want to know which models are supported by NUT before purchasing one. In both of those cases, more information about how USB IDs map to models can help smooth out those processes. At the moment, we manually add each protocol number to the usbhid-ups driver when a user tries an UPS that isn't listed already. If there were a convention that all USB idDevice values in a certain range were going to be HID PDC compliant, we could change the default from opt-in to enabled-by-default (but we wouldn't want the UPS driver to try to control a USB hub).”
>>  
> Let me kick this around. If we were able to say: USB UPSes with IDs from “09ae:XXXX” to 09ae:YYYY” are PDC compliant, would that be enough?
> Topically the problem I can think of is determining if this holds for older units (I’ll dig into this).

I think a range of IDs (even with a few exceptions for the older units) would be sufficient. The ideal scenario is that the range is somewhat future-proof, so a version of NUT from this year can properly identify next year's UPS. If not, we still have manual ways for users to add their idProduct to ups.conf and the udev files, but as you can imagine, that is frustrating for a new user.