https://bugs.kde.org/show_bug.cgi?id=402480
--- Comment #2 from Khem Raj <raj.k...@gmail.com> --- (In reply to Mark Wielaard from comment #1) > Comment on attachment 117070 [details] > Patch to fix the issue > > While I agree that trying to clobber the stack pointer doesn't make sense > (and it only "works" because we never return from the asm), the changing of > the stack pointer itself is deliberate and part of the test (bug #345887 was > about checking the stack pointer was bogus and not crashing valgrind, even > though the test program itself does crash). > > Also without the bogus stack pointer the expected output doesn't match > anymore (because now we get a full backtrace because we can follow the > stack). > > So how about just removing the clobber, but keeping stack pointer change > itself: > > diff --git a/none/tests/amd64-linux/bug345887.c > b/none/tests/amd64-linux/bug345887.c > index 0f9237d..269bd70 100644 > --- a/none/tests/amd64-linux/bug345887.c > +++ b/none/tests/amd64-linux/bug345887.c > @@ -20,13 +20,17 @@ static void inner(void) > "movq $0x10d, %%r14\n" > "movq $0x10e, %%r15\n" > // not %rbp as mdb is then not able to reconstruct stack trace > + // Do change %rsp (to test a bogus stack pointer), > + // but don't add %rsp to the clobber list since gcc ignores it > + // and since gcc >= 9.0 errors about it > + // see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=52813 > "movq $0x10f, %%rsp\n" > "movq $0x1234, (%%rax)\n" // should cause SEGV here > "ud2" // should never get here > : // no output registers > : // no input registers > : "memory", "%rax", "%rbx", "%rcx", "%rdx", "%rsi", "%rdi", > - "%r8", "%r9", "%r10", "%r11", "%r12", "%r13", "%r14", "%r15", > "%rsp"); > + "%r8", "%r9", "%r10", "%r11", "%r12", "%r13", "%r14", "%r15"); > } > > __attribute__((noinline)) OK that seems fine to me. It will work. -- You are receiving this mail because: You are watching all bug changes.
