diff mbox series

[RFC,v5,19/28] dyndbg: RFC handle __dyndbg* sections in module.lds.h

Message ID 20210511185057.3815777-20-jim.cromie@gmail.com (mailing list archive)
State New
Headers show
Series dynamic debug diet plan | expand

Commit Message

Jim Cromie May 11, 2021, 6:50 p.m. UTC
Up until now, loadable modules have tacitly linked the 2 __dyndbg*
sections into the .ko, as observable in log by enabling module.c's
pr_debugs and loading a module.

But now, we need to explicitly place the header sections in front of
their respective __dyndbg* sections, so we reuse input section names
for the output.

This gives us the placement we need for the header record, which we
can see in the "add-module:"s and elements "0 0" below:

    "0 0" lines are headers: predicate (function==module && !lineno)
    "X debug prints in" are 1 too high, they count headers.
    we are adding tables for empty modules (1st 2 below)

[    7.578873] dyndbg: add-module: ghash_clmulni_intel
[    7.579716] dyndbg:  0 0 ghash_clmulni_intel.ghash_clmulni_intel.0
[    7.608995] dyndbg:   1 debug prints in module ghash_clmulni_intel
[    8.078085] dyndbg: add-module: rapl
[    8.078977] dyndbg:  0 0 rapl.rapl.0
[    8.079584] dyndbg:   1 debug prints in module rapl
[    8.082009] RAPL PMU: API unit is 2^-32 Joules, 0 fixed counters, 10737418240 ms ovfl timer
[    8.099294] dyndbg: add-module: intel_rapl_common
[    8.100177] dyndbg:  0 0 intel_rapl_common.intel_rapl_common.0
[    8.101026] dyndbg:  1 1 intel_rapl_common.rapl_remove_package.1279
[    8.101931] dyndbg:  2 2 intel_rapl_common.rapl_detect_domains.1245
[    8.102836] dyndbg:  3 3 intel_rapl_common.rapl_detect_domains.1242
[    8.103778] dyndbg:  4 4 intel_rapl_common.rapl_package_register_powercap.1159
[    8.104960] dyndbg:  5 5 intel_rapl_common.rapl_package_register_powercap.1145
[    8.106246] dyndbg:  6 6 intel_rapl_common.rapl_package_register_powercap.1114
[    8.107302] dyndbg:  7 7 intel_rapl_common.rapl_package_register_powercap.1108
[    8.108338] dyndbg:  8 8 intel_rapl_common.rapl_update_domain_data.1083
[    8.109278] dyndbg:  9 9 intel_rapl_common.rapl_check_unit_atom.824
[    8.110255] dyndbg:  10 10 intel_rapl_common.rapl_check_unit_core.796
[    8.111361] dyndbg:  11 11 intel_rapl_common.rapl_read_data_raw.722
[    8.112301] dyndbg:  12 12 intel_rapl_common.contraint_to_pl.303
[    8.113276] dyndbg:  13 debug prints in module intel_rapl_common
[    8.130452] dyndbg: add-module: intel_rapl_msr
[    8.131140] dyndbg:  0 0 intel_rapl_msr.intel_rapl_msr.0
[    8.132026] dyndbg:  1 1 intel_rapl_msr.rapl_msr_probe.172
[    8.132818] dyndbg:  2 2 intel_rapl_msr.rapl_msr_read_raw.104
[    8.133794] dyndbg:   3 debug prints in module intel_rapl_msr

This gives us the property we need:

   fixed offset from &__dyndbg[N] to &__dyndbg[0]
   container_of gets &header
   header has ptr to __dyndbg_sites[]
   we can (in principle) drop __dyndbg.site ptr
   (after we adapt header to keep it)

TODO:

At this point, for loaded modules, ddebug_add_module() sees the header
as 0'th element, as we need in order to drop site (and regain worst
case footprint parity).

It could/should properly init this header to support the _sites[n]
lookup for loaded mods.  Or at least handle it explicitly.  Or at
least see what proc-show does with it currently.

Note that for builtins, decided by (__start <= dp < __stop), we use
__start___dyndbg_sites[N] directly, and dont need the header.

But maybe we should use it anyway, double-checking/BUGing when wrong.

Signed-off-by: Jim Cromie <jim.cromie@gmail.com>
---
 include/asm-generic/module.lds.h | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)
diff mbox series

Patch

diff --git a/include/asm-generic/module.lds.h b/include/asm-generic/module.lds.h
index f210d5c1b78b..328c48dfc88c 100644
--- a/include/asm-generic/module.lds.h
+++ b/include/asm-generic/module.lds.h
@@ -4,7 +4,17 @@ 
 
 /*
  * <asm/module.lds.h> can specify arch-specific sections for linking modules.
- * Empty for the asm-generic header.
+ *
+ * For loadable modules with CONFIG_DYNAMIC_DEBUG, we need to keep the
+ * 2 __dyndbg* ELF sections, which are loaded by module.c
+ *
+ * Pack the 2 __dyndbg* input sections with their respective
+ * .gnu.linkonce. header records into 2 output sections, with those
+ * header records in the 0th element.
  */
+SECTIONS {
+__dyndbg_sites	: ALIGN(8) { *(.gnu.linkonce.dyndbg_site) *(__dyndbg_sites) }
+__dyndbg	: ALIGN(8) { *(.gnu.linkonce.dyndbg)	  *(__dyndbg) }
+}
 
 #endif /* __ASM_GENERIC_MODULE_LDS_H */