On 5/3/26 2:05 AM, Jim Cromie wrote: > Several build configs had problems with __dyndbg* sections getting > lost in drm drivers. Fix this by following the model demonstrated in > codetag.lds.h. > > Introduce include/asm-generic/dyndbg.lds.h, to bundle dynamic-debug's > multiple sections together, into 2 macros: > > vmlinux.lds.h DATA_DATA: move the 2 BOUNDED_SECTION_BY(__dyndbg*) > calls into dyndbg.lds.h DYNDBG_SECTIONS(). vmlinux.lds.h now includes > the new file and calls the new macro. > > MOD_DYNDBG_SECTIONS declares the 2 BOUNDED_SECTION_BY calls, but wraps > them with output section syntax to keep them as known and separate ELF > sections in the module.ko. The KEEP fixes the lost section. > > dyndbg.lds.h includes (reuses) bounded-section.lds.h > > scripts/module.lds.S: now calls MOD_DYNDBG_SECTIONS right before the > CODETAG macro (consistent with their placements in vmlinux.lds.h), and > also includes dyndbg.lds.h > > This isolates vmlinux.lds.h from further __dyndbg section additions. > > CC: Arnd Bergmann <[email protected]> > Signed-off-by: Jim Cromie <[email protected]> > --- [...] > diff --git a/include/asm-generic/dyndbg.lds.h > b/include/asm-generic/dyndbg.lds.h > new file mode 100644 > index 000000000000..f95683aa16b6 > --- /dev/null > +++ b/include/asm-generic/dyndbg.lds.h > @@ -0,0 +1,19 @@ > +/* SPDX-License-Identifier: GPL-2.0-only */ > +#ifndef __ASM_GENERIC_DYNDBG_LDS_H > +#define __ASM_GENERIC_DYNDBG_LDS_H > + > +#include <asm-generic/bounded_sections.lds.h> > +#define DYNDBG_SECTIONS() \ > + . = ALIGN(8); \ > + BOUNDED_SECTION_BY(__dyndbg, ___dyndbg) \ > + BOUNDED_SECTION_BY(__dyndbg_classes, ___dyndbg_classes) > + > +#define MOD_DYNDBG_SECTIONS() \ > + __dyndbg : { \ > + BOUNDED_SECTION_BY(__dyndbg, ___dyndbg) \ > + } \ > + __dyndbg_classes : { \ > + BOUNDED_SECTION_BY(__dyndbg_classes, ___dyndbg_classes) \ > + }
The BOUNDED_SECTION_BY() macro always defines the __start and __end symbols for the section. This is not ideal for modules because these symbols are created even when no __dyndbg or __dyndbg_classes input sections are present. Additionally, they cause MOD_DYNDBG_SECTIONS() to produce dummy output sections with the same names in all modules, even when dynamic debug is disabled in the config. This unnecessarily pollutes the modules. Since modules are relocatable files, the usual method for locating consolidated data in a module is to read its section table. The second problem is that MOD_DYNDBG_SECTIONS() should force the address of all its output sections to 0. Otherwise, when linking modules with 'ld.bfd -r', sections defined without an address inherit the location counter, resulting in non-zero sh_addr values in the resulting .ko files. Non-zero addresses are confusing in this context, typically worse compressible, and may cause external tools to misbehave. -- Thanks, Petr
