[PATCH v2 1/5] drm/ttm: Update Makefile for KUnit

Christian König christian.koenig at amd.com
Mon Sep 11 13:42:58 UTC 2023



Am 11.09.23 um 15:33 schrieb Karolina Stolarek:
> On 11.09.2023 15:12, Christian König wrote:
>> Am 11.09.23 um 13:47 schrieb Karolina Stolarek:
>>> On 11.09.2023 12:49, Jani Nikula wrote:
>>>> On Mon, 11 Sep 2023, Karolina Stolarek 
>>>> <karolina.stolarek at intel.com> wrote:
>>>>> Update Makefile so it can produce a module that consists of TTM 
>>>>> tests.
>>>>> This will allow us to test non-exported functions when KUnit tests
>>>>> are built as a module. Remove the tests' Makefile.
>>>>
>>>> I'm asking questions instead of making assertions, because I'm not 
>>>> 100%
>>>> confident, but I don't feel like this Makefile could work right.
>>>
>>> Questions, assertions and comments are fine, I'm glad you're taking 
>>> a look at the patch :) I'm not 100% confident either, so I welcome 
>>> suggestions on how to improve it.
>>>
>>> The problem is that TTM tests try to test symbols that are not 
>>> exported, so I have to compile all the files together into one 
>>> module if I choose to build KUnit tests as a module. The other 
>>> option that I'm considering is to make the tests are builtin only. 
>>> I'm tempted to go with it (for the sake of simplicity), but I'm 
>>> trying to get the module option to work first.
>>
>> I have to admit that this looks pretty awkward, but I'm not an expert 
>> on the Linux build system in the first place.
>
> Neither am I, and it shows :)
>
>> Would it be an option to build the unit tests into the standard 
>> ttm.ko module and let them run whenever that module loads?
>
> You mean appending the list of tests to ttm-y, depending on 
> $(CONFIG_DRM_TTM_KUNIT_TEST)?

Yes.

> I _think_ I tried something similar, and couldn't get it to work, got 
> a bunch of "twice exported" warnings.

You might need to adjust module_init, and things like MODULE_VERSION 
etc..., but I would give it a try.

Thanks for looking into this,
Christian.

>
>> On the other hand if this solution here works, why not?
>
> Because it's complicated and, well, awkward. I'm still thinking about 
> a use case where we would prefer to have KUnit tests defined as a 
> module. IIRC, in CI we enable KUnit tests as bultins and run them 
> during the boot. kunit.py helper also defines the tests as builtins.
>
> All the best,
> Karolina
>
>>
>> Regards,
>> Christian.
>>
>>>
>>>>
>>>>>
>>>>> Signed-off-by: Karolina Stolarek <karolina.stolarek at intel.com>
>>>>> Reported-by: kernel test robot <lkp at intel.com>
>>>>> Closes: 
>>>>> https://lore.kernel.org/oe-kbuild-all/202309010358.50gYLkmw-lkp@intel.com/
>>>>> Closes: 
>>>>> https://lore.kernel.org/oe-kbuild-all/202309011134.bwvpuyOj-lkp@intel.com/
>>>>> Closes: 
>>>>> https://lore.kernel.org/oe-kbuild-all/202309011935.bBpezbUQ-lkp@intel.com/
>>>>> ---
>>>>>   drivers/gpu/drm/ttm/Makefile       | 18 +++++++++++++-----
>>>>>   drivers/gpu/drm/ttm/tests/Makefile |  6 ------
>>>>>   2 files changed, 13 insertions(+), 11 deletions(-)
>>>>>   delete mode 100644 drivers/gpu/drm/ttm/tests/Makefile
>>>>>
>>>>> diff --git a/drivers/gpu/drm/ttm/Makefile 
>>>>> b/drivers/gpu/drm/ttm/Makefile
>>>>> index dad298127226..6322a33e65ed 100644
>>>>> --- a/drivers/gpu/drm/ttm/Makefile
>>>>> +++ b/drivers/gpu/drm/ttm/Makefile
>>>>> @@ -2,10 +2,18 @@
>>>>>   #
>>>>>   # Makefile for the drm device driver.  This driver provides 
>>>>> support for the
>>>>>   -ttm-y := ttm_tt.o ttm_bo.o ttm_bo_util.o ttm_bo_vm.o 
>>>>> ttm_module.o \
>>>>> -    ttm_execbuf_util.o ttm_range_manager.o ttm_resource.o 
>>>>> ttm_pool.o \
>>>>> -    ttm_device.o ttm_sys_manager.o
>>>>> +ttm := ttm_tt.o ttm_bo.o ttm_bo_util.o ttm_bo_vm.o ttm_module.o \
>>>>> +       ttm_execbuf_util.o ttm_range_manager.o ttm_resource.o 
>>>>> ttm_pool.o \
>>>>> +       ttm_device.o ttm_sys_manager.o
>>>>> +obj-$(CONFIG_DRM_TTM) += $(ttm)
>>>>
>>>> Does that not lead to each object in $(ttm) becoming its own module?
>>>
>>> Huh, yes, that is what would happen here. Not good...
>>>
>>>>
>>>>>   ttm-$(CONFIG_AGP) += ttm_agp_backend.o
>>>>
>>>> Does this not create a ttm.o with just one object, depending on
>>>> CONFIG_AGP?
>>>
>>> I just left it as it was before my changes.
>>>
>>>>
>>>>>   -obj-$(CONFIG_DRM_TTM) += ttm.o
>>>>> -obj-$(CONFIG_DRM_TTM_KUNIT_TEST) += tests/
>>>>> +ttm-tests := tests/ttm_kunit_helpers.o tests/ttm_device_test.o \
>>>>> +             tests/ttm_pool_test.o
>>>>
>>>> I'd preserve the one object per line syntax. It's nicer for the 
>>>> diffs in
>>>> subsequent updates.
>>>
>>> OK, will update it in the next version (if such comes). I just 
>>> followed the style of "ttm-y".
>>>
>>>>
>>>>> +
>>>>> +ifeq ($(CONFIG_DRM_TTM_KUNIT_TEST),m)
>>>>> +    ttm-test-objs := $(ttm) $(ttm-tests)
>>>>
>>>> Isn't the -objs syntax for host/userspace programs? And if not, 
>>>> doesn't
>>>> it lead to objects with exported symbols being present in two places?
>>>
>>> I saw it in use in other Makefiles, so I followed it. As for the 
>>> exported symbols, I tested both builtin and module configs, and it 
>>> was fine, but it's possible I missed something. I suspect that the 
>>> variables are first expanded, and then processed by the Makefile.
>>>
>>> Many thanks,
>>> Karolina
>>>
>>>>
>>>> Confused.
>>>>
>>>> BR,
>>>> Jani.
>>>>
>>>>> +    obj-m := ttm-test.o
>>>>> +else
>>>>> +    obj-$(CONFIG_DRM_TTM_KUNIT_TEST) += $(ttm-tests)
>>>>> +endif
>>>>> diff --git a/drivers/gpu/drm/ttm/tests/Makefile 
>>>>> b/drivers/gpu/drm/ttm/tests/Makefile
>>>>> deleted file mode 100644
>>>>> index ec87c4fc1ad5..000000000000
>>>>> --- a/drivers/gpu/drm/ttm/tests/Makefile
>>>>> +++ /dev/null
>>>>> @@ -1,6 +0,0 @@
>>>>> -# SPDX-License-Identifier: GPL-2.0 AND MIT
>>>>> -
>>>>> -obj-$(CONFIG_DRM_TTM_KUNIT_TEST) += \
>>>>> -        ttm_device_test.o \
>>>>> -        ttm_pool_test.o \
>>>>> -        ttm_kunit_helpers.o
>>>>
>>



More information about the dri-devel mailing list