Hello,
kselftest can be built using the kernel's top most Makefile without using kselftest's Makefile directly. But there is bug in the top most Makefile. The build fails if the specified output directory is first level sub-directory. Here is a example to reproduce this bug:
make kselftest-all O=build
"The Make is working in a wrong directory, that is why the relative path does not work." Masahiro Yamada. Feel free to fix it if someone pin the bug.
It should be noted that the build works in some other combinations: make kselftest-all (works) make kselftest-all O=/tmp (works) make kselftest-all O=build/build2 (works)
My unsuccessful attempt to fix this bug can be found here: https://lore.kernel.org/lkml/20220223191016.1658728-1-usama.anjum@collabora....
Thanks, Muhammad Usama Anjum
On Sat, May 21, 2022 at 3:44 PM Muhammad Usama Anjum usama.anjum@collabora.com wrote:
Hello,
kselftest can be built using the kernel's top most Makefile without using kselftest's Makefile directly. But there is bug in the top most Makefile. The build fails if the specified output directory is first level sub-directory. Here is a example to reproduce this bug:
make kselftest-all O=build
"The Make is working in a wrong directory, that is why the relative path does not work." Masahiro Yamada. Feel free to fix it if someone pin the bug.
It should be noted that the build works in some other combinations: make kselftest-all (works) make kselftest-all O=/tmp (works) make kselftest-all O=build/build2 (works)
My unsuccessful attempt to fix this bug can be found here: https://lore.kernel.org/lkml/20220223191016.1658728-1-usama.anjum@collabora....
Thanks, Muhammad Usama Anjum
This problem starts from the bad design of the kselftest framework. I did some research before. I think I can fix the root cause but currently I do not have enough time to do it.
KBUILD_ABS_SRCTREE is a user-interface to request Kbuild to use the absolute path. If it is forced in the top Makefile, users have no way to negate it. It is true that using the absolute path is a quick work-around because you do not need to care about the current working directory.
If you insist on it, just go ahead. It is just two line changes. Once the issue is fixed in a better way, your patch can be reverted easily.
On 5/22/22 10:15 PM, Masahiro Yamada wrote:
On Sat, May 21, 2022 at 3:44 PM Muhammad Usama Anjum usama.anjum@collabora.com wrote:
Hello,
kselftest can be built using the kernel's top most Makefile without using kselftest's Makefile directly. But there is bug in the top most Makefile. The build fails if the specified output directory is first level sub-directory. Here is a example to reproduce this bug:
make kselftest-all O=build
"The Make is working in a wrong directory, that is why the relative path does not work." Masahiro Yamada. Feel free to fix it if someone pin the bug.
It should be noted that the build works in some other combinations: make kselftest-all (works) make kselftest-all O=/tmp (works) make kselftest-all O=build/build2 (works)
My unsuccessful attempt to fix this bug can be found here: https://lore.kernel.org/lkml/20220223191016.1658728-1-usama.anjum@collabora....
Thanks, Muhammad Usama Anjum
This problem starts from the bad design of the kselftest framework. I did some research before. I think I can fix the root cause but currently I do not have enough time to do it.
KBUILD_ABS_SRCTREE is a user-interface to request Kbuild to use the absolute path. If it is forced in the top Makefile, users have no way to negate it. It is true that using the absolute path is a quick work-around because you do not need to care about the current working directory.
If you insist on it, just go ahead. It is just two line changes. Once the issue is fixed in a better way, your patch can be reverted easily.
Thank you for replying. Lets merge it Shuah?
Hi Shuah,
Should we merge this?
Thanks, Usama
On 5/23/22 2:40 PM, Muhammad Usama Anjum wrote:
On 5/22/22 10:15 PM, Masahiro Yamada wrote:
On Sat, May 21, 2022 at 3:44 PM Muhammad Usama Anjum usama.anjum@collabora.com wrote:
Hello,
kselftest can be built using the kernel's top most Makefile without using kselftest's Makefile directly. But there is bug in the top most Makefile. The build fails if the specified output directory is first level sub-directory. Here is a example to reproduce this bug:
make kselftest-all O=build
"The Make is working in a wrong directory, that is why the relative path does not work." Masahiro Yamada. Feel free to fix it if someone pin the bug.
It should be noted that the build works in some other combinations: make kselftest-all (works) make kselftest-all O=/tmp (works) make kselftest-all O=build/build2 (works)
My unsuccessful attempt to fix this bug can be found here: https://lore.kernel.org/lkml/20220223191016.1658728-1-usama.anjum@collabora....
Thanks, Muhammad Usama Anjum
This problem starts from the bad design of the kselftest framework. I did some research before. I think I can fix the root cause but currently I do not have enough time to do it.
KBUILD_ABS_SRCTREE is a user-interface to request Kbuild to use the absolute path. If it is forced in the top Makefile, users have no way to negate it. It is true that using the absolute path is a quick work-around because you do not need to care about the current working directory.
If you insist on it, just go ahead. It is just two line changes. Once the issue is fixed in a better way, your patch can be reverted easily.
Thank you for replying. Lets merge it Shuah?
On 5/22/22 11:15 AM, Masahiro Yamada wrote:
On Sat, May 21, 2022 at 3:44 PM Muhammad Usama Anjum usama.anjum@collabora.com wrote:
Hello,
kselftest can be built using the kernel's top most Makefile without using kselftest's Makefile directly. But there is bug in the top most Makefile. The build fails if the specified output directory is first level sub-directory. Here is a example to reproduce this bug:
make kselftest-all O=build
"The Make is working in a wrong directory, that is why the relative path does not work." Masahiro Yamada. Feel free to fix it if someone pin the bug.
It should be noted that the build works in some other combinations: make kselftest-all (works) make kselftest-all O=/tmp (works) make kselftest-all O=build/build2 (works)
My unsuccessful attempt to fix this bug can be found here: https://lore.kernel.org/lkml/20220223191016.1658728-1-usama.anjum@collabora....
Thanks, Muhammad Usama Anjum
This problem starts from the bad design of the kselftest framework. I did some research before. I think I can fix the root cause but currently I do not have enough time to do it.
KBUILD_ABS_SRCTREE is a user-interface to request Kbuild to use the absolute path. If it is forced in the top Makefile, users have no way to negate it. It is true that using the absolute path is a quick work-around because you do not need to care about the current working directory.
If you insist on it, just go ahead. It is just two line changes. Once the issue is fixed in a better way, your patch can be reverted easily.
Why don't we work on fixing it the wright way? I would rather go that route than using short rem fixes.
Usama, would you be interested in working on a proper fix as recommended by Masahiro?
thanks, -- Shuah
This problem starts from the bad design of the kselftest framework. I did some research before. I think I can fix the root cause but currently I do not have enough time to do it.
KBUILD_ABS_SRCTREE is a user-interface to request Kbuild to use the absolute path. If it is forced in the top Makefile, users have no way to negate it. It is true that using the absolute path is a quick work-around because you do not need to care about the current working directory.
If you insist on it, just go ahead. It is just two line changes. Once the issue is fixed in a better way, your patch can be reverted easily.
Why don't we work on fixing it the wright way? I would rather go that route than using short rem fixes.
Usama, would you be interested in working on a proper fix as recommended by Masahiro?
I've made another attempt: https://lore.kernel.org/lkml/20220627070214.432390-1-usama.anjum@collabora.c...
thanks, -- Shuah
linux-kselftest-mirror@lists.linaro.org
-
Masahiro Yamada -
Muhammad Usama Anjum -
Shuah Khan