|
From: Andreas A. <ar...@li...> - 2019-04-08 11:08:53
|
On Fri, Apr 05 2019, Julian Seward wrote: > https://sourceware.org/git/gitweb.cgi?p=valgrind.git;h=d36ea889d8d8a1646be… > > commit d36ea889d8d8a1646be85c30ab5771af6912b7a1 > Author: Julian Seward <jse...@pu...> > Date: Fri Apr 5 20:10:46 2019 +0200 > > Bug 404843 - s390x: backtrace sometimes ends prematurely. Thanks again for taking care of this. > [...] > --- a/coregrind/m_libcassert.c > +++ b/coregrind/m_libcassert.c > @@ -163,19 +163,38 @@ > [...] > + "std %%f5, 64(%1);" \ > + "std %%f6, 72(%1);" \ > + "std %%f7, 80(%1);" \ > + : /* out */ "=r" (ia) \ > + : /* in */ "r" (&block[0]) \ This "r" should be an "a" instead. For "r" the compiler could theoretically choose %r0, which wouldn't do what we want. The "=r" above is OK, because "brasl" works with %r0 as well. > + : /* trash */ "memory" \ Fine by me. Normally I prefer to be more precise (as shown in my proposal), but I guess it doesn't matter here. -- Andreas |