Hi Jiaming, Thanks for the patch.
The change correctly moves the remaining foreground probes into isolated temporary build directories and keeps them tracked by the top-level AC_KERNEL_WAIT. Wrapping each probe individually also avoids the nested-background PID tracking issue. This prevents stale conftest.o state from affecting the checks and makes these probes consistent with the rest. Reviewed-by: Bob Zhou <[email protected]> We will apply the patch. Regards, Bob ________________________________________ From: Jiaming Fan Sent: Tuesday, 10 November 2026 01:44:10 To: Chengjun Yao; Zhou, Bob; Su, Yang Cc: Russell, Kent; Cui, Flora; [email protected] Subject: [PATCH] drm/amdkcl: run the remaining foreground probes in background [Some people who received this message don't often get email from [email protected]. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ] The HDMI and DP colorspace property probes and the legacy gamma probe are the only top-level probes that are not wrapped in AC_KERNEL_DO_BACKGROUND. They run in the shared configure directory, one at a time, while every other probe runs in parallel in its own temporary directory. Running in the shared directory also lets a probe see a conftest.o left there by an earlier one. Before commit a4e55d0e6d9a ("drm/amdkcl: string objtool from cflags for conftest") the probe command ended in "; objtool ... conftest.o", so a failed compile was hidden by objtool succeeding on that stale object. On v5.17 this defined HAVE_DRM_MODE_CREATE_{HDMI,DP}_COLORSPACE_PROPERTY_2ARGS for kernels that only have the 1-argument functions. That commit removed the cause; running these probes in background like the others removes the shared state as well. Wrap each probe separately rather than the colorspace group as a whole: AC_AMDGPU_DRM_CONNECT_ATTACH_COLORSPACE_PROPERTY already runs in background, and a job started from inside another one is not in $procs, so AC_KERNEL_WAIT would not wait for it. The generated config.h is unchanged. Link: https://github.com/ROCm/amdgpu/issues/228 Signed-off-by: Jiaming Fan --- This was a real bug in the ROCm 6.0.0 DKMS package on kernel v5.17, which I reported in issue #228. It is already fixed on master by a4e55d0e6d9a, so this patch does not fix a bug there; it only makes these three probes consistent with the rest. Feel free to drop it if you prefer to leave them as they are. ...rm_mode_create_hdmi_colorspace_property.m4 | 40 ++++++++++--------- .../drm/amd/dkms/m4/struct_drm_crtc_funcs.m4 | 12 +++--- 2 files changed, 29 insertions(+), 23 deletions(-) diff --git a/drivers/gpu/drm/amd/dkms/m4/drm_mode_create_hdmi_colorspace_property.m4 b/drivers/gpu/drm/amd/dkms/m4/drm_mode_create_hdmi_colorspace_property.m4 index ecc33db..3d7eee7 100644 --- a/drivers/gpu/drm/amd/dkms/m4/drm_mode_create_hdmi_colorspace_property.m4 +++ b/drivers/gpu/drm/amd/dkms/m4/drm_mode_create_hdmi_colorspace_property.m4 @@ -20,30 +20,34 @@ dnl # commit v6.1-5783-g08383039cd19 dnl # drm/connector: Allow drivers to pass list of supported colorspaces dnl # AC_DEFUN([AC_AMDGPU_DRM_MODE_CREATE_HDMI_COLORSPACE_PROPERTY], [ - AC_KERNEL_TRY_COMPILE_SYMBOL([ - #include - ], [ - drm_mode_create_hdmi_colorspace_property(NULL, 0); - ], [drm_mode_create_hdmi_colorspace_property], [drivers/gpu/drm/drm_connector.c], [ - AC_DEFINE(HAVE_DRM_MODE_CREATE_HDMI_COLORSPACE_PROPERTY_2ARGS, 1, - [drm_mode_create_hdmi_colorspace_property() has 2 args]) - ]) -]) + AC_KERNEL_DO_BACKGROUND([ + AC_KERNEL_TRY_COMPILE_SYMBOL([ + #include + ], [ + drm_mode_create_hdmi_colorspace_property(NULL, 0); + ], [drm_mode_create_hdmi_colorspace_property], [drivers/gpu/drm/drm_connector.c], [ + AC_DEFINE(HAVE_DRM_MODE_CREATE_HDMI_COLORSPACE_PROPERTY_2ARGS, 1, + [drm_mode_create_hdmi_colorspace_property() has 2 args]) + ]) + ]) +]) dnl # dnl # commit v6.1-5783-g08383039cd19 dnl # drm/connector: Allow drivers to pass list of supported colorspaces dnl # AC_DEFUN([AC_AMDGPU_DRM_MODE_CREATE_DP_COLORSPACE_PROPERTY], [ - AC_KERNEL_TRY_COMPILE_SYMBOL([ - #include - ], [ - drm_mode_create_dp_colorspace_property(NULL, 0); - ], [drm_mode_create_dp_colorspace_property], [drivers/gpu/drm/drm_connector.c], [ - AC_DEFINE(HAVE_DRM_MODE_CREATE_DP_COLORSPACE_PROPERTY_2ARGS, 1, - [drm_mode_create_dp_colorspace_property() has 2 args]) - ]) -]) + AC_KERNEL_DO_BACKGROUND([ + AC_KERNEL_TRY_COMPILE_SYMBOL([ + #include + ], [ + drm_mode_create_dp_colorspace_property(NULL, 0); + ], [drm_mode_create_dp_colorspace_property], [drivers/gpu/drm/drm_connector.c], [ + AC_DEFINE(HAVE_DRM_MODE_CREATE_DP_COLORSPACE_PROPERTY_2ARGS, 1, + [drm_mode_create_dp_colorspace_property() has 2 args]) + ]) + ]) +]) AC_DEFUN([AC_AMDGPU_DRM_MODE_CREATE_COLORSPACE_PROPERTY_FUNCS], [ diff --git a/drivers/gpu/drm/amd/dkms/m4/struct_drm_crtc_funcs.m4 b/drivers/gpu/drm/amd/dkms/m4/struct_drm_crtc_funcs.m4 index 6b53674..9e9acbc 100644 --- a/drivers/gpu/drm/amd/dkms/m4/struct_drm_crtc_funcs.m4 +++ b/drivers/gpu/drm/amd/dkms/m4/struct_drm_crtc_funcs.m4 @@ -23,13 +23,15 @@ dnl # dnl # v5.10-1961-g6ca2ab8086af drm: automatic legacy gamma support dnl # AC_DEFUN([AC_AMDGPU_STRUCT_DRM_CRTC_FUNCS_GAMMA_SET_OPTIONAL], [ - AC_KERNEL_TRY_COMPILE_SYMBOL([ - #include - ], [ - drm_atomic_helper_legacy_gamma_set(NULL, NULL, NULL, NULL, 0, NULL); - ], [drm_atomic_helper_legacy_gamma_set], [drivers/gpu/drm/drm_atomic_helper.c],[],[ + AC_KERNEL_DO_BACKGROUND([ + AC_KERNEL_TRY_COMPILE_SYMBOL([ + #include + ], [ + drm_atomic_helper_legacy_gamma_set(NULL, NULL, NULL, NULL, 0, NULL); + ], [drm_atomic_helper_legacy_gamma_set], [drivers/gpu/drm/drm_atomic_helper.c],[],[ AC_DEFINE(HAVE_STRUCT_DRM_CRTC_FUNCS_GAMMA_SET_OPTIONAL, 1, [HAVE_STRUCT_DRM_CRTC_FUNCS_GAMMA_SET_OPTIONAL is available]) + ]) ]) ]) -- 2.56.0
