On Mon, Feb 24, 2025 at 10:54:13AM -0700, Simon Glass wrote:
Hi Tom,
On Fri, 21 Feb 2025 at 09:06, Tom Rini <tr...@konsulko.com> wrote:
On Fri, Feb 21, 2025 at 06:57:34AM -0700, Simon Glass wrote:
Hi Tom,
On Thu, 20 Feb 2025 at 07:53, Tom Rini <tr...@konsulko.com> wrote:
On Thu, Feb 20, 2025 at 06:49:49AM -0700, Simon Glass wrote:
Hi Tom,
On Tue, 18 Feb 2025 at 17:55, Tom Rini <tr...@konsulko.com> wrote:
On Tue, Feb 18, 2025 at 05:01:40PM -0700, Simon Glass wrote:
Hi Tom,
On Tue, 18 Feb 2025 at 08:11, Tom Rini <tr...@konsulko.com> wrote:
On Tue, Feb 18, 2025 at 05:09:23AM -0700, Simon Glass wrote:
Hi Tom,
On Mon, 17 Feb 2025 at 10:52, Tom Rini <tr...@konsulko.com> wrote:
On Sun, Feb 16, 2025 at 01:44:13PM -0700, Simon Glass wrote:
Now that U-Boot can boot this quickly, using kvm, add a test that the
installer starts up correctly.
Use the qemu-x86_64 board in the SJG lab.
Signed-off-by: Simon Glass <s...@chromium.org>
---
Changes in v2:
- Add more patches to support booting with kvm
- Add new patch with a test for booting Ubuntu 24.04
.gitlab-ci.yml | 5 ++++
test/py/tests/test_distro.py | 53 ++++++++++++++++++++++++++++++++++++
2 files changed, 58 insertions(+)
create mode 100644 test/py/tests/test_distro.py
diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index 8c49d5b0a79..ec799e97c10 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -745,3 +745,8 @@ zybo:
variables:
ROLE: zybo
<<: *lab_dfn
+
+qemu-x86_64:
+ variables:
+ ROLE: qemu-x86_64
+ <<: *lab_dfn
I'm not sure why this is in your lab stanza, rather than the normal
test.py QEMU stanza.
Are you wanting to add the Ubuntu image into CI? It is quite large.
If we're going to be able to run it on N platforms, yes, we need to
think of a good way to cache the download. There's not a particular
reason we can't run the stock Ubuntu RISC-V image on the two sifive
targets and also qemu-riscv64, is there?
Yes, we can do that. It is pretty simple to set up in Labgrid and it
doesn't require all the runners to download a much larger image, etc.
I don't quite understand why it's under "labgrid". These are generic CI
tests. Now maybe we need to, in both Gitlab and Azure, add some logic so
that certain longer or possibly destructive tests are only run on tagged
releases or as requested rather than every time, as it will take longer.
But pretty much every platform under the qemu target list should be able
to Just Boot an off the shelf OS distribution is my point.
Sure, and I'm not suggesting we shouldn't do that as well.
diff --git a/test/py/tests/test_distro.py b/test/py/tests/test_distro.py
new file mode 100644
index 00000000000..51eec45cecc
--- /dev/null
+++ b/test/py/tests/test_distro.py
@@ -0,0 +1,53 @@
+# SPDX-License-Identifier: GPL-2.0+
+# Copyright 2025 Canonical Ltd.
+# Written by Simon Glass <simon.gl...@canonical.com>
+
+import pytest
+
+DOWN = '\x1b\x5b\x42\x0d'
+
+# Enable early console so that the test can see if something goes wrong
+CONSOLE = 'earlycon=uart8250,io,0x3f8 console=uart8250,io,0x3f8'
+
+@pytest.mark.boardspec('qemu-x86_64')
+@pytest.mark.role('qemu-x86_64')
+def test_distro(ubman):
+ """Test that of-platdata can be generated and used in sandbox"""
+ with ubman.log.section('boot'):
+ ubman.run_command('boot', wait_for_prompt=False)
+
+ with ubman.log.section('Grub'):
+ # Wait for grub to come up and offset a menu
+ ubman.p.expect(['Try or Install Ubuntu'])
+
+ # Press 'e' to edit the command line
+ ubman.run_command('e', wait_for_prompt=False, send_nl=False)
+
+ # Wait until we see the editor appear
+ ubman.p.expect(['/casper/initrd'])
+
+ # Go down to the 'linux' line
+ ubman.send(DOWN * 3)
+
+ # Go to end of line
+ ubman.ctrl('E')
+
+ # Backspace to remove 'quiet splash'
+ ubman.send('\b' * len('quiet splash'))
+
+ # Send our noisy console
+ ubman.send(CONSOLE)
+
+ # Tell grub to boot
+ ubman.ctrl('X')
+ ubman.p.expect(['Booting a command list'])
+
+ with ubman.log.section('Linux'):
+ # Linux should start immediately
+ ubman.p.expect(['Linux version'])
+
+ with ubman.log.section('Ubuntu'):
+ # Shortly later, we should see this banner
+ ubman.p.expect(['Welcome to .*Ubuntu 24.04.1 LTS.*!'])
+
+ ubman.restart_uboot()
And this seems very inflexible. Please see
test/py/tests/test_net_boot.py for an example of how to have this be
configurable and work on arbitrary platforms. What I assume is tricky is
that the "role" part here is where you have a special disk image being
passed. That too could be dealt with in u-boot-test-hooks in a few ways,
and the images pre-fetched to the CI container. And if this was
configurable similar to the example I noted above, it could check real
hardware too.
That wasn't the reaction I expected.
Yes, it is inflexible, but it is a starting point. Isn't it better
than what we have today?
Is your inflexible boot an OS test better than the flexible boot an OS
test that we have today? No, it's not.
I didn't even know about it, or perhaps I forgot.
I believe I mentioned it every time you've said we should have an OS
test, so yes, I guess you forgot.
Well it was only added in May last year and it relies on board config
which I don't have...although I see that you have now posted yours.
Yes, it was added not quite a year ago, and is documented within the
test, like most tests that rely on the real platform.
And do we need better documentation for test? Yes.
+1
I'll note that I did my bit!
Perhaps this relates to getting the labgrid config published and
figuring out how to pass info from Labgrid to tests.
I would like to generalise this test to work on at least one real
board, preferably one that doesn't use grub.
OK. The test we have today does that, if you check for the "Welcome to
..." string instead of the kernel has booted string. It also does
netboot rather than run default bootcmd. But that's an easy enough test
to write up. The only thing stopping me from doing that right now is I
need to find a board in the lab where we installed an OS to eMMC and not
SD card (some lab sd-mux issues).
OK. Labgrid has a 'features' thing which you can attach to targets, so
I should be able to use that to indicate that Ubuntu, Debian, Armbian,
etc. are available.
OK, but that sounds like the opposite direction. These are generic tests
that can run in any / all of the labs, not just your labgrid
configuration. AMD has been contributing tests that run on hardware for
example.
That's great, the more tests we have the better. But those tests can't
and don't run in CI, whereas mine can and do.
AFAICT they're running on AMD's CI. They run on my CI. They don't run on
*your* lab because you took things, intentionally, in a direction to
minimize using u-boot-test-hooks and our existing per-board
configuration infrastructure.
When I look at CI all I see is my lab. Which CI are you referring to
and how can I access it?
I'll point you at the notes for the first call we had recently:
https://lore.kernel.org/u-boot/20250128171923.GQ1233568@bill-the-cat/
and note that there are many labs doing testing on / with U-Boot.
That's all good, but it isn't as good as having the lab in gitlab.
Strongly disagree. Especially since having it in the mainline gitlab
isn't feasible.
Here I would like to make a case for moving to using Labgrid across
the board, but unfortunately the project struggles to review PRs, so
it's probably not a good idea.
It would also be counter to the feedback from the U-Boot community about
making it easier to contribute testing results from additional labs.
I really don't think the test hooks are a good setup, though. It is
OK-ish for small labs, but it is so fiddly to use that I wrote a tool
(Labman) to deal with all the confusion.
Yes, I don't know how hard you evaluated all of the then-current lab
management tooling and wrote your own.
Labgrid (which you suggested I use for my lab, if you recall),
Yes, and I think you forgot the aim was to make it easy to show all of
the existing Labgrid based labs that do Linux kernel testing they could
easily add U-Boot to the mix. I've been trying to get feedback from
other people with existing labgrid setups to look at what you've done.
provides for two yaml configuration files so that everything is in one
place. Apart from its primitive support for USB hubs, it is much
easier to maintain that dozens of little files all over the palce.
I mean, I looked at what you posted and strongly disagree, but I think
both cases here are personal preference and not some sort of objective
and easily evaluated thing.
We need an 'all of the above' strategy here.
Sure. But I still want to see things as reusable as possible. What you
have above is *extremely* board and OS specific and non-configurable.
Yes, agreed.
I
also don't quite see why it's not a test of autoboot with the
pre-requisite of an OS being installed.
Ah OK, my test is just for the installer itself. Both are useful, but
I hope eventually to have the installer run to completion and then
reboot to check all is well.
In the spirit of "yes, and.."'ing tests, sure. Ilias pointed me at some
testing Linaro has going now that automates I believe it was current
Yocto and current U-Boot (+ the pmb patches that've been posted) doing a
full install via network in CI. So yes, a Canonical lab might also find
it useful to end to end test installing Ubuntu. My own personal dream is
that at least some of the existing kernelci labs see the utility in
adding "current U-Boot" as one of the matrix variables they test and not
just "U-Boot as delivered by vendor" as a static part of the testing.
OK.
BTW, having thought about how test/py works a bit, instead of the
env__net_tftp_bootable_file stuff, we should have code or data which
sets up the required test files (on a suitable server) before running
the test. That way, all the test code is in one Python file and we
don't have to spend ages trying to divine what each test needs.
That seems like a lot more work than documenting more what we have
today, and I'm not sure of the benefit. Given the contents of the pxe
test, yes, just having those files available to 'cp' in place would be
helpful. But that's not the case for booting a kernel (the FIT match
stuff doesn't work on the TI platforms atm). And if you look at the
config I posted it also includes bootstage configuration. It also won't
work well for the SPI tests, which I'm talking with Love about in
another thread.
Yes, perhaps, but having self-contained tests would be a win.
With it's own set of technical and legal challenges / obligations and
difficulties depending on what you even mean by "self contained". And
how often what's run where, and all sorts of other challenges too.
Given the extreme depth that testing can go to, this is why I'm of the
position that we need to document things more and worry less about
prepackaged things. For example, making the documentation for the
current net based OS boot means that for bringing up a new board the
developer can just drop something in. Whereas if the tests expect a
functional OS image that has to also be messed with and is its own
challenge.
Yes
In other words, the majority of py/<host>/u_boot_boardenv_ content is
configuration details, specific to both the platform / SoC first, some
lab specific details second and drop-in existing 3rd party files a
distant third.
I think the u-boot-test-hooks was an amazing solution 9 years ago, but
we have outgrown it. We want people to be able to connect their lab to
CI (meaning gitlab), so testing is more automated.
More and more public testing would be great. The notes I linked above
explain one of the first problems there being that most companies will
not or can not hook a lab to a public CI instance.
Well corporate IT is what it is.
That means that their boards will not be testing in CI, unless they do
it themselves, right?
I'm not sure what you mean here. It's a solved problem for them (monitor
tree at URL) and something that's been being done since the beginning
even for U-Boot (it's how the original nvidia lab worked).
The next problem, as
both of our personal labs show, is that just maintaining the physical
lab takes time and resources. I've added Heiko here because I've been
talking with him off-list about expanding tbot coverage and plumbing
that in to gitlab.
OK
We should move away from relying on maintainers getting around to
testing patches months after they are sent, when they have time, but
they don't. Things need to be more automated and I'd encourage you to
push this as well.
I have been, and the results I've gotten are that companies are testing
things internally but there's not any good way to publish results, and
that's the kind of framework we're entirely missing.
If you like, but from my side, I like to see the results in gitlab.
Depends on what you mean by gitlab. I assume you mean "triggered by a
push and visible in the main pipeline". Which isn't possible. It's not
going to happen. External collection is how it's handled for the linux
kernel and that community has far more sway than we do. If we ride their
coattails here so to speak, we can get results. If we push for something
completely different we aren't likely to have success.
Which is another part of why I keep pushing against having U-Boot
configuration stuff inside of Labgrid as it makes it harder for any lab
that's not using labgrid to see how to configure things.
Well, as you requested, I looked at Labgrid and now my lab uses it. I
am happy to publish the config[1], but I still hold my view that all
the shell scripts in u-boot-test-hooks are limiting and painful to
work with.
Yes, and as I've shown, you can also use labgrid without going down the
same path you took, and we can also support other lab management methods
too, which is important to get as much testing as possible without
needing to centralize everything.
