| Message ID | 20220829055223.24767-2-sv@linux.ibm.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | objtool: Enable and implement --mcount option on powerpc | expand |
Le 29/08/2022 à 07:52, Sathvika Vasireddy a écrit : > objtool is throwing *unannotated intra-function call* warnings in > .c files with a few instructions that are marked unreachable. The > problem comes from the annotate_unreachable() macro that is > called by unreachable(). This annotation is adding a call to a > function with size 0, and objtool does not add such symbols > to the rbtree. Due to this reason, find_call_destination() function > is not able to find the destination symbol for that call. > > With the annotation (annotate_unreachable()), gcc seems to > generate a 'bl' to unreachable symbol with size 0. But with > the builtin variant of unreachable (__builtin_unreachable()), > gcc does not emit calls to such symbols and the warnings > go away. Given that the codegen remains same, and that > there are no 'bl' instructions to such symbols emitted, fix > these warnings by replacing unreachable() with it's builtin > variant in __WARN_FLAGS(). How can you say that the codegen remains the same if with the original you get stale 'bl' instructions and with the alternative you don't ? > > Also, add barrier_before_unreachable() before __builtin_unreachable() > to work around a gcc bug [1], for the problem reported at [2]. Here my comment was not related to the gcc bug [1] but to gcc bug https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82365 , which was worked around by commit 173a3efd3edb ("bug.h: work around GCC PR82365 in BUG()") By chance it also solve the problem [1] as you mention. > > [1]: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106751 > [2]: https://lkml.org/lkml/2022/8/25/418 > > Signed-off-by: Sathvika Vasireddy <sv@linux.ibm.com> Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu> > --- > arch/powerpc/include/asm/bug.h | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/arch/powerpc/include/asm/bug.h b/arch/powerpc/include/asm/bug.h > index 61a4736355c2..ef42adb44aa3 100644 > --- a/arch/powerpc/include/asm/bug.h > +++ b/arch/powerpc/include/asm/bug.h > @@ -99,7 +99,8 @@ > __label__ __label_warn_on; \ > \ > WARN_ENTRY("twi 31, 0, 0", BUGFLAG_WARNING | (flags), __label_warn_on); \ > - unreachable(); \ > + barrier_before_unreachable(); \ > + __builtin_unreachable(); \ > \ > __label_warn_on: \ > break; \
Christophe Leroy wrote: > > > Le 29/08/2022 à 07:52, Sathvika Vasireddy a écrit : >> objtool is throwing *unannotated intra-function call* warnings in >> .c files with a few instructions that are marked unreachable. The >> problem comes from the annotate_unreachable() macro that is >> called by unreachable(). This annotation is adding a call to a >> function with size 0, and objtool does not add such symbols >> to the rbtree. Due to this reason, find_call_destination() function >> is not able to find the destination symbol for that call. >> >> With the annotation (annotate_unreachable()), gcc seems to >> generate a 'bl' to unreachable symbol with size 0. But with >> the builtin variant of unreachable (__builtin_unreachable()), >> gcc does not emit calls to such symbols and the warnings >> go away. Given that the codegen remains same, and that >> there are no 'bl' instructions to such symbols emitted, fix >> these warnings by replacing unreachable() with it's builtin >> variant in __WARN_FLAGS(). > > How can you say that the codegen remains the same if with the original > you get stale 'bl' instructions and with the alternative you don't ? I guess the reference to codegen remaining the same is more to do with unreachable vs. __builtin_unreachable() in the absence of CONFIG_OBJTOOL. But yeah, the changelog needs to be reworked to clarify that. > >> >> Also, add barrier_before_unreachable() before __builtin_unreachable() >> to work around a gcc bug [1], for the problem reported at [2]. > > Here my comment was not related to the gcc bug [1] but to gcc bug > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82365 , which was worked > around by commit 173a3efd3edb ("bug.h: work around GCC PR82365 in BUG()") > > By chance it also solve the problem [1] as you mention. That's a good commit to reference, but please also retain a link to the new PR. - Naveen
diff --git a/arch/powerpc/include/asm/bug.h b/arch/powerpc/include/asm/bug.h index 61a4736355c2..ef42adb44aa3 100644 --- a/arch/powerpc/include/asm/bug.h +++ b/arch/powerpc/include/asm/bug.h @@ -99,7 +99,8 @@ __label__ __label_warn_on; \ \ WARN_ENTRY("twi 31, 0, 0", BUGFLAG_WARNING | (flags), __label_warn_on); \ - unreachable(); \ + barrier_before_unreachable(); \ + __builtin_unreachable(); \ \ __label_warn_on: \ break; \
objtool is throwing *unannotated intra-function call* warnings in .c files with a few instructions that are marked unreachable. The problem comes from the annotate_unreachable() macro that is called by unreachable(). This annotation is adding a call to a function with size 0, and objtool does not add such symbols to the rbtree. Due to this reason, find_call_destination() function is not able to find the destination symbol for that call. With the annotation (annotate_unreachable()), gcc seems to generate a 'bl' to unreachable symbol with size 0. But with the builtin variant of unreachable (__builtin_unreachable()), gcc does not emit calls to such symbols and the warnings go away. Given that the codegen remains same, and that there are no 'bl' instructions to such symbols emitted, fix these warnings by replacing unreachable() with it's builtin variant in __WARN_FLAGS(). Also, add barrier_before_unreachable() before __builtin_unreachable() to work around a gcc bug [1], for the problem reported at [2]. [1]: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106751 [2]: https://lkml.org/lkml/2022/8/25/418 Signed-off-by: Sathvika Vasireddy <sv@linux.ibm.com> --- arch/powerpc/include/asm/bug.h | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)