Hi Tom,
On Fri, 30 Aug 2024 at 09:13, Tom Rini trini@konsulko.com wrote:
On Thu, Aug 29, 2024 at 07:09:22PM -0600, Simon Glass wrote:
Hi Tom,
On Thu, 29 Aug 2024 at 09:20, Tom Rini trini@konsulko.com wrote:
On Thu, Aug 29, 2024 at 09:02:38AM -0600, Simon Glass wrote:
Hi Neil,
On Thu, 29 Aug 2024 at 08:44, neil.armstrong@linaro.org wrote:
On 29/08/2024 14:17, Simon Glass wrote:
Hi Peter,
On Thu, 29 Aug 2024 at 04:43, Peter Robinson pbrobinson@gmail.com wrote: > > On Wed, 28 Aug 2024 at 22:25, Simon Glass sjg@chromium.org wrote: >> >> Hi Peter, >> >> On Wed, 28 Aug 2024 at 12:14, Peter Robinson pbrobinson@gmail.com wrote: >>> >>> Hi Simon, >>> >>>> With Labgrid we don't need to specify the various methods, except for >>>> the console, which simply calls labgrid-client. >>>> >>>> This allows supporting any boards in your lab, without adding per-board >>>> configuration to these hooks. >>>> >>>> Provide ellesmere files as an example. >>> >>> What's ellesmere? >> >> It is a lake but also the name of a computer. >> >>> >>>> Signed-off-by: Simon Glass sjg@chromium.org >>>> --- >>>> >>>> Changes in v4: >>>> - Support pytest fully with dual-build boards like Beagleplay >>>> >>>> Changes in v3: >>>> - Update scripts for latest version of Labgrid integration >>>> - Add poweroff.none and poweron.none >>>> - Provide -n flag when querying board info >>>> - Target the grpc version of Labgrid which is now in -master >>>> - Update README to cover the changes >>>> >>>> Changes in v2: >>>> - Make use of the common script (only) to set bin_dir >>>> >>>> README.md | 50 ++++++++++++++++++++++++++++++++++++ >>> >>> Maybe that should be in a separate labsgrid readme? >> >> My hope is that Labgrid becomes the normal way of running these tests, > > Generally I agree with automated testing platforms, I think <insert > you're preferred platform here> makes sense, there's a bunch like > Linaro and a bunch in the arm ecosystem use LAVA [1] and there's a > bunch more that I'm aware of.
I am somewhat familiar with LAVA and I believe it can be used to test U-Boot, although I need to learn how. Looking at a test run [2] for beaglebone black I see that it is using a recent kernel but the U-Boot seems to be older.
> > Does it make sense, of course at some point in the future post this > being merged, make sense to look at a general way of making it easier > to plugin these sort of HW test platforms using this as a basis? I ask > mostly because putting a bunch of my devices into some sort of > platform can auto test things and of course everyone has an opinion to > which is the best one :-P
Yes. I had heard from Tom that Labgrid is the new hotness for now. Having dug into it I believe it is a good solution, although it can certainly be improved to handle scale better.
Anyway, IMO the current test hooks are not a great solution, just because the configuration is spread all over the place and it relies on lots of little shell scripts. So I believe that the Labgrid integration is a closer to where we want to be with others that come along.
I'd say all those scripts are actually here to ease integration with any system, booting U-Boot and Linux are two different beasts.
That's fine, go ahead and use the scripts. My point is that Labgrid doesn't need them and in fact it makes everything pretty painful if we try to use all of them.
I guess I really need to clean-up and post my former co-workers scripts as I strongly disagree with that statement.
See some examples below. Bear in mind also that my goal has been to get my lab fully running, that includes interactive access to boards, as well as running tests. For that I have been using tbot (with integrated build and software-loading features). WIth Labgrid I have been able to replace most of the pytest scripts, tbot and Labman with Labgrid + 60 patches and some configuration files.
That's certainly good and useful progress. And I feel like one of the lessons of this thread is that every lab ends up a bit different and so there's just not going to be as much standardization as we'd all like.
Let's look through the scripts:
u-boot-test-common - new script
u-boot-test-console - needed
u-boot-test-flash - here's an example:
. poweroff.${power_impl} sleep 0.1 . flash.sdwire_common_mount . poweron.${power_impl}
Here's another:
# Handles the common SDwire mounting (caller does power control)
mount_dir=/media/${mount_point}
# Switch over to get USB card access sd-mux-ctrl --device-serial ${sdwire_serial} --ts
complete=false for i in {0..9}; do if out="$(mount UUID=${mount_uuid} 2>&1)"; then complete=true break fi echo $out
# If it is already mounted, try to unmount it first. It may have been # mounted by another user so we won't have the access we need. If this gives # an error then we know we cannot continue if [[ $out == *"already mounted"* ]]; then umount UUID=${mount_uuid} fi sleep 1done if [[ $complete = false ]]; then echo "Failed to mount UUID ${mount_uuid} after 10 tries" exit 1 fi
# Sanity check if ! mountpoint -q ${mount_dir}; then echo "Mount ${mount_dir} not available after 'mount'" exit 1 fi
# Perform the write, pass along as much environment as possible . writer.${flash_writer}
complete=false for i in {0..9}; do if out="$(umount ${mount_dir} 2>&1)"; then complete=true break fi echo $out sleep 1 done
if [[ $complete = false ]]; then echo "Failed to umount UUID ${mount_uuid} after 10 tries" exit 1 fi
# Sanity check if mountpoint -q ${mount_dir}; then echo "Mount ${mount_dir} still available after 'umount'" exit 1 fi
# Back to card access for the DUT sd-mux-ctrl --device-serial ${sdwire_serial} --dut
Of course that assumes an SDwire which may not always be the case.
Or some other piece of hardware that sd-mux-ctrl supports, but yes.
Abstracting all of this into shell scripts is just too painful.
Abstracting it to something else sounds even more painful however.
I see this a little bit like the situation with Binman. We now have a description for the image and can build for any SoC which has added its peculiarities in a Python script.
In a similar way, I believe it is valuable to have each SoC encode (in some sort of description) how it writes images to an SD card, over USB, etc.
Now I am not a huge fan of yaml but it does the job and it is what Labgrid uses. So my vision of all this is that if someone wants to set up a lab with 5 boards, they can get the gear together, wire it all up, then copy and modify some yaml and it should all work.
Also I have found it to be error-prone.
There is also bin/flash.sdwire_relay_mount:
. poweroff.${power_impl}
sleep 0.1
. flash.sdwire_common_mount
. poweron.${power_impl}
Yup. It's possible we could do some slightly nicer check/sleep loop but it probably wouldn't really get us much.
u-boot-test-getrole
- New script
Which is for how you abstracted hardware from software configurations it can run (for example, the Pi 4 case of rpi_4, rpi_4_32b and rpi_arm64 all being valid defconfigs).
Yes it maps a 'role' (like an 'identity' in test.py) to a particular DUT and the corresponding U-Boot board. It can be many-to-one.
u-boot-test-power-off
- This one fits with Labgrid quite well, but of course there may not
actually be power control
u-boot-test-power-on
- This one also fits with Labgrid, but since serial output can come
out as soon as it is called, jumping back into Labgrid later to bring up a console results in lots of missing output. Also there may not actually be power control
Right, but in both cases, poweron/poweroff.labgrid could just do: labgrid-client power $state || true if power isn't supported (which sounds odd, but, maybe makes sense in some setup).
Yes it can. I have quite a few boards with no power control (they are just on all the time). I find a reset line more useful than power in some cases.
u-boot-test-release
- New script. Actually this isn't really needed, since I added
automatic release to Labgrid. I was just worried that Labgrid might crash
What I saw in working on my scripts was that if you don't release you can't acquire later and there's not an option for "you already have acquired it". On my scripts, I should update the cleanup function to release the current LG_PLACE.
Yes I hit the same problem. I added an '--auto' option to acquire and release, so they are more forgiving.
u-boot-test-reset
- Unfortunately labgrid doesn't provide a reset option, although I'm
sure it could be added. You still need to know whether to use power or reset or both, though, so again you have shell scripts with this stuff
Perhaps this is something to talk with upstream about, for the cases where "reset" is not just another name for "power cycle" ? I know we have scripts today for a number of cases but the only ones I wrote are just power cycle (digital-loggers) or power cycle via USB (ykush, the USB hubs that support per-port power control intentionally).
If you did down a bit it ends up being intertwined with the U-Boot strategy...e.g. you for USB you often need to hold reset down while holding recovery, then release reset, then write U-Boot and wait for it to come up.
reset.recovery_download - another thing I had to add, BTW
. "${bin_dir}/recovery.${recovery_impl}"
# USB enumeration delay for ((i = 0; i <= 100; i++)); do if [ -e "${board_usb_dev}" ]; then break fi sleep 0.1 done if [ ! -e "${board_usb_dev}" ]; then echo "Cannot find device ${board_usb_dev}" exit 1 fi sleep 1
. "${bin_dir}/download.${download_impl}"
or another case, this is Zynq which tries to wait a bit before calling picocom. It mostly works, but it loses output
# Some boards have a UART built into them and it doesn't appear until the # board is powered on. An example is zynq_zybo. Try to handle this by waiting # for the console and then continuing as soon as possible (so we don't miss # the U-Boot banner). if [ -n "${console_wait}" ]; then for i in {0..99}; do if [ -e "${console_dev}" ]; then break fi sleep .1 done if [ ! -e "${console_dev}" ]; then echo "Console ${console_dev} not found" exit 1 fi fi
exec picocom -b "${console_baud}" "${console_dev}"
So that's about a third of the reason why I don't think the scripts scale.
This is where I think looking at https://github.com/bootlin/snagboot which Kevin Hillman pointed me at a while ago might come in handy, instead.
Oh yes I ran into that at a show a year or two back, very nice! Better than having a lot of little tools.
The only "painful" part is the shared pain, regardless of framework, for "put blob $X at location $Y", which I believe you abstract out in to python?
Sort-of. The Python knows how to handle all the different writing methods and is at [2], but it is entirely controlled by yaml[3] (the example_env.yaml). See the ti,am625 method for example and the 'play' board right at the bottom of the end file.
I very much like the configuration files (although I hope one day Labgrid will allow the two files to be combined). It makes it easier to add a new board, since there is much less to think about. It really did take a long time to fiddle with all those hook scripts and I got tired of it.
Regards, Simon
[1] https://github.com/sjg20/u-boot/tree/lab6a/tools/labman [2] https://github.com/labgrid-project/labgrid/pull/1411/commits/c7ea03987362af7... [3] https://github.com/labgrid-project/labgrid/pull/1411/commits/6d3b4a0cfc7e35e...
Yeah, personally I don't care for that. I know I'm not going to convince people and it's just one of those "goodness, I'm the old guy now" things, but, no. It would be easier if you had comments for each SoC as to what / why each of the magic values, but then shouldn't it be abstracted more? Perhaps to a file per SoC, and at that point how is this easier to follow than a commented shell script? But in sum, I'm just cranky about yaml and I know your opinion is the opposite.
As above I don't like it either, but it is what Labgrid has chosen. In fact I used it with Labman too...it really needs proper phandles like devicetree. At least is isn't JSON though...:-)
As an aside, please review the ti,am625 stuff with some TI folks, that's both far too specific (you're going to need to add different am62 and am64 and am69 and j7... and on and on cases) and I forget if there's some work to have a slightly easier way to know what file to use going on.
Well, perhaps they can take it over and get everything working? I have done the hard yards, e.g. supporting two builds together.
Regards, Simon