On Sun, Nov 23, 2025 at 04:25:51PM +0100, René Rebe wrote:
> Hey,
> 
> On Sun, 23 Nov 2025 16:09:41 +0100 (CET), René Rebe <[email protected]> wrote:
> 
> > On Sun, 23 Nov 2025 15:57:24 +0100, "Christophe Leroy (CS GROUP)" 
> > <[email protected]> wrote:
> > 
> > > Le 23/11/2025 à 13:13, René Rebe a écrit :
> > > > Building a size optimized ppc64 kernel (-Os), gcc emits more FP
> > > > save/restore symbols, that the linker generates on demand into the
> > > > .sfpr section. Explicitly allow-list those in scripts/mod/modpost.c,
> > > > too. They are needed for the amdgpu in-kernel floating point support.
> > > 
> > > Would have been interested to know with which version of GCC the
> > > problem started.
> > 
> > idk, maybe forever, or at least a decade fo GCC? Most devs probably
> > don't build size optimized, and addtionally we only use in kernel
> > floating point for amdgpu since recently? Should I add Fixes: for the
> > in-kernel FP hash?
> > 
> > > By the way you seem to fix the problem for modules, but does it also
> > > work when amdgpu is in kernel ? I would have expected a need to add
> > > functions in arch/powerpc/lib/crtsavres.S as well, just like following
> > > commits:
> > > 
> > > 8fe9c93e7453 ("powerpc: Add vr save/restore functions")
> > > 7fca5dc8aa7a ("powerpc: Fix module building for gcc 4.5 and 64 bit")
> > > da3de6df33f5 ("[POWERPC] Fix -Os kernel builds with newer gcc
> > > versions")
> > 
> > idk, I avoid linking that big stuff directly into the kernel and would
> > need to specically test that, too. I guess I go do that now, too, ...
> 
> It appears built-in amdgpu FP somehow magically works for me:
> 
> debug-linux:[linux-6.17]# grep DRM.*AMD .config
> CONFIG_DRM_AMDGPU=y
> CONFIG_DRM_AMDGPU_SI=y
> CONFIG_DRM_AMDGPU_CIK=y
> CONFIG_DRM_AMDGPU_USERPTR=y
> CONFIG_DRM_AMD_ACP=y
> CONFIG_DRM_AMD_DC=y
> CONFIG_DRM_AMD_DC_FP=y
> CONFIG_DRM_AMD_DC_SI=y
> ...
>   CC      drivers/gpu/drm/amd/amdgpu/../display/modules/hdcp/hdcp_ddc.o
>   CC      drivers/gpu/drm/amd/amdgpu/../display/modules/hdcp/hdcp_log.o
>   CC      drivers/gpu/drm/amd/amdgpu/../display/modules/hdcp/hdcp_psp.o
>   CC      drivers/gpu/drm/amd/amdgpu/../display/modules/hdcp/hdcp.o
>   CC      drivers/gpu/drm/amd/amdgpu/../display/modules/hdcp/hdcp1_execution.o
>   CC      
> drivers/gpu/drm/amd/amdgpu/../display/modules/hdcp/hdcp1_transition.o
>   CC      drivers/gpu/drm/amd/amdgpu/../display/modules/hdcp/hdcp2_execution.o
>   CC      
> drivers/gpu/drm/amd/amdgpu/../display/modules/hdcp/hdcp2_transition.o
>   AR      drivers/gpu/drm/amd/amdgpu/built-in.a
>   AR      drivers/gpu/drm/built-in.a
>   AR      drivers/gpu/built-in.a
>   AR      drivers/built-in.a
>   AR      built-in.a
>   AR      vmlinux.a
>   LD      vmlinux.o
>   GEN     modules.builtin.modinfo
>   GEN     modules.builtin
>   MODPOST vmlinux.symvers
>   CC      .vmlinux.export.o
>   UPD     include/generated/utsversion.h
>   CC      init/version-timestamp.o
>   KSYMS   .tmp_vmlinux0.kallsyms.S
>   AS      .tmp_vmlinux0.kallsyms.o
>   LD      .tmp_vmlinux1
>   NM      .tmp_vmlinux1.syms
>   KSYMS   .tmp_vmlinux1.kallsyms.S
>   AS      .tmp_vmlinux1.kallsyms.o
>   LD      .tmp_vmlinux2
>   NM      .tmp_vmlinux2.syms
>   KSYMS   .tmp_vmlinux2.kallsyms.S
>   AS      .tmp_vmlinux2.kallsyms.o
>   LD      vmlinux.unstripped
>   NM      System.map
>   SORTTAB vmlinux.unstripped
> make[3]: Nothing to be done for 'vmlinux.unstripped'.
>   OBJCOPY vmlinux
> 
> So I guess the patch is good to go after clarifying which kind of
> Fixes: to use?

Was this ever picked up or addressed elswhere?

> > > > MODPOST Module.symvers
> > > > ERROR: modpost: "_restfpr_20" [drivers/gpu/drm/amd/amdgpu/amdgpu.ko]
> > > > undefined!
> > > > ERROR: modpost: "_restfpr_26" [drivers/gpu/drm/amd/amdgpu/amdgpu.ko]
> > > > undefined!
> > > > ERROR: modpost: "_restfpr_22" [drivers/gpu/drm/amd/amdgpu/amdgpu.ko]
> > > > undefined!
> > > > ERROR: modpost: "_savegpr1_27" [drivers/gpu/drm/amd/amdgpu/amdgpu.ko]
> > > > undefined!
> > > > ERROR: modpost: "_savegpr1_25" [drivers/gpu/drm/amd/amdgpu/amdgpu.ko]
> > > > undefined!
> > > > ERROR: modpost: "_restfpr_28" [drivers/gpu/drm/amd/amdgpu/amdgpu.ko]
> > > > undefined!
> > > > ERROR: modpost: "_savegpr1_29" [drivers/gpu/drm/amd/amdgpu/amdgpu.ko]
> > > > undefined!
> > > > ERROR: modpost: "_savefpr_20" [drivers/gpu/drm/amd/amdgpu/amdgpu.ko]
> > > > undefined!
> > > > ERROR: modpost: "_savefpr_22" [drivers/gpu/drm/amd/amdgpu/amdgpu.ko]
> > > > undefined!
> > > > ERROR: modpost: "_restfpr_15" [drivers/gpu/drm/amd/amdgpu/amdgpu.ko]
> > > > undefined!
> > > > WARNING: modpost: suppressed 56 unresolved symbol warnings because
> > > > there were too many)
> > > > Signed-off-by: René Rebe <[email protected]>
> > > > ---
> > > > V2: description
> > > > Theoretically for -stable, but no previous commit that broke it.
> > > 
> > > In that case you have to add Cc: [email protected]
> > > Add indeed it is likely a gcc upgrade that broke it, not a previous
> > > commit.
> > 
> > Should I then simply use enabling amdgpu dc_fp and in-kernel FP as the
> > breaking commit for Fixes:?
> > 
> > Thanks!
> > 
> >     René
> > 
> > > > ---
> > > >   scripts/mod/modpost.c | 4 ++++
> > > >   1 file changed, 4 insertions(+)
> > > > diff --git a/scripts/mod/modpost.c b/scripts/mod/modpost.c
> > > > index 47c8aa2a6939..133dfa16308a 100644
> > > > --- a/scripts/mod/modpost.c
> > > > +++ b/scripts/mod/modpost.c
> > > > @@ -602,6 +602,10 @@ static int ignore_undef_symbol(struct elf_info
> > > > *info, const char *symname)
> > > >                 /* Special register function linked on all modules 
> > > > during final link of
> > > >                 .ko */
> > > >                 if (strstarts(symname, "_restgpr0_") ||
> > > >                     strstarts(symname, "_savegpr0_") ||
> > > > +                   strstarts(symname, "_restgpr1_") ||
> > > > +                   strstarts(symname, "_savegpr1_") ||
> > > > +                   strstarts(symname, "_restfpr_") ||
> > > > +                   strstarts(symname, "_savefpr_") ||
> > > >                     strstarts(symname, "_restvr_") ||
> > > >                     strstarts(symname, "_savevr_") ||
> > > >                     strcmp(symname, ".TOC.") == 0)
> > > 
> > 
> > -- 
> > René Rebe, ExactCODE GmbH, Berlin, Germany
> > https://exactco.dehttps://t2linux.comhttps://patreon.com/renerebe
> 
> -- 
> René Rebe, ExactCODE GmbH, Berlin, Germany
> https://exactco.dehttps://t2linux.comhttps://patreon.com/renerebe

Reply via email to