2022-11-11 22:39:42

by Jim Cromie

Subject: [PATCH 1/7] drm: mark drm.debug-on-dyndbg as BROKEN for now

drm.debug-on-dyndbg has a regression, due to a chicken-egg
initialization problem:

1- modprobe i915
i915 needs drm.ko, which is loaded 1st

2- "modprobe drm drm.debug=0x1ff" (virtual/implied)
drm.debug is set post-initialization, from boot-args etc

3- `modprobe i915` finishes

W/O drm.debug-on-dyndbg that just works, because all drm_dbg*
callsites use drm_debug_enabled() to check __drm_debug & DEM_UT_<CAT>
before printing.

But the whole point of drm.debug-on-dyndbg is to avoid that runtime
test, by enabling (at post-modinit) a static-key at each callsite in
the just-loaded module.

And since drm.ko is loaded before all dependent modules, none are
"just-loaded", and no drm.debug callsites are present yet, except
those in drm.ko itself.

Signed-off-by: Jim Cromie <[email protected]>
---
drivers/gpu/drm/Kconfig | 1 +
1 file changed, 1 insertion(+)

diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig
index 34f5a092c99e..0d1e59e6bb7e 100644
--- a/drivers/gpu/drm/Kconfig
+++ b/drivers/gpu/drm/Kconfig
@@ -54,6 +54,7 @@ config DRM_DEBUG_MM
config DRM_USE_DYNAMIC_DEBUG
bool "use dynamic debug to implement drm.debug"
default y
+ depends on BROKEN # chicken-egg initial enable problem
depends on DRM
depends on DYNAMIC_DEBUG || DYNAMIC_DEBUG_CORE
depends on JUMP_LABEL
--
2.38.1

2022-11-14 13:22:12

by Ville Syrjälä

Subject: Re: [Intel-gfx] [PATCH 1/7] drm: mark drm.debug-on-dyndbg as BROKEN for now

On Fri, Nov 11, 2022 at 03:17:09PM -0700, Jim Cromie wrote:
> drm.debug-on-dyndbg has a regression, due to a chicken-egg
> initialization problem:
>
> 1- modprobe i915
> i915 needs drm.ko, which is loaded 1st
>
> 2- "modprobe drm drm.debug=0x1ff" (virtual/implied)
> drm.debug is set post-initialization, from boot-args etc
>
> 3- `modprobe i915` finishes
>
> W/O drm.debug-on-dyndbg that just works, because all drm_dbg*
> callsites use drm_debug_enabled() to check __drm_debug & DEM_UT_<CAT>
> before printing.
>
> But the whole point of drm.debug-on-dyndbg is to avoid that runtime
> test, by enabling (at post-modinit) a static-key at each callsite in
> the just-loaded module.
>
> And since drm.ko is loaded before all dependent modules, none are
> "just-loaded", and no drm.debug callsites are present yet, except
> those in drm.ko itself.
>
> Signed-off-by: Jim Cromie <[email protected]>
> ---
> drivers/gpu/drm/Kconfig | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig
> index 34f5a092c99e..0d1e59e6bb7e 100644
> --- a/drivers/gpu/drm/Kconfig
> +++ b/drivers/gpu/drm/Kconfig
> @@ -54,6 +54,7 @@ config DRM_DEBUG_MM
> config DRM_USE_DYNAMIC_DEBUG
> bool "use dynamic debug to implement drm.debug"
> default y

Could you switch to 'default n' as well? i915 CI actually enables
BROKEN so that we can test some more experimental stuff which is
hidden behind BROKEN for normal users.

> + depends on BROKEN # chicken-egg initial enable problem
> depends on DRM
> depends on DYNAMIC_DEBUG || DYNAMIC_DEBUG_CORE
> depends on JUMP_LABEL
> --
> 2.38.1

--
Ville Syrj?l?
Intel

2022-11-17 06:53:53

by Greg Kroah-Hartman

Subject: Re: [PATCH 1/7] drm: mark drm.debug-on-dyndbg as BROKEN for now

On Fri, Nov 11, 2022 at 03:17:09PM -0700, Jim Cromie wrote:
> drm.debug-on-dyndbg has a regression, due to a chicken-egg
> initialization problem:
>
> 1- modprobe i915
> i915 needs drm.ko, which is loaded 1st
>
> 2- "modprobe drm drm.debug=0x1ff" (virtual/implied)
> drm.debug is set post-initialization, from boot-args etc
>
> 3- `modprobe i915` finishes
>
> W/O drm.debug-on-dyndbg that just works, because all drm_dbg*
> callsites use drm_debug_enabled() to check __drm_debug & DEM_UT_<CAT>
> before printing.
>
> But the whole point of drm.debug-on-dyndbg is to avoid that runtime
> test, by enabling (at post-modinit) a static-key at each callsite in
> the just-loaded module.
>
> And since drm.ko is loaded before all dependent modules, none are
> "just-loaded", and no drm.debug callsites are present yet, except
> those in drm.ko itself.
>
> Signed-off-by: Jim Cromie <[email protected]>
> ---
> drivers/gpu/drm/Kconfig | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/drivers/gpu/drm/Kconfig b/drivers/gpu/drm/Kconfig
> index 34f5a092c99e..0d1e59e6bb7e 100644
> --- a/drivers/gpu/drm/Kconfig
> +++ b/drivers/gpu/drm/Kconfig
> @@ -54,6 +54,7 @@ config DRM_DEBUG_MM
> config DRM_USE_DYNAMIC_DEBUG
> bool "use dynamic debug to implement drm.debug"
> default y
> + depends on BROKEN # chicken-egg initial enable problem
> depends on DRM
> depends on DYNAMIC_DEBUG || DYNAMIC_DEBUG_CORE
> depends on JUMP_LABEL
> --
> 2.38.1

This should go through the drm tree now. The rest probably should also
go that way and not through my tree as well.

thanks,

greg k-h