Re: [PATCH 1/2] KVM: selftests: Make rseq compatible with glibc-2.35

[Gavin Shan]

Hi Florian,

On 8/9/22 4:33 PM, Florian Weimer wrote:
> > diff --git a/tools/testing/selftests/kvm/rseq_test.c b/tools/testing/selftests/kvm/rseq_test.c
> > index a54d4d05a058..acb1bf1f06b3 100644
> > --- a/tools/testing/selftests/kvm/rseq_test.c
> > +++ b/tools/testing/selftests/kvm/rseq_test.c
> > @@ -9,6 +9,7 @@
> >   #include <string.h>
> >   #include <signal.h>
> >   #include <syscall.h>
> > +#include <dlfcn.h>
> >   #include <sys/ioctl.h>
> >   #include <sys/sysinfo.h>
> >   #include <asm/barrier.h>
>
> I'm surprised that there isn't a Makefile update to link with -ldl
> (still required for glibc 2.33 and earlier).

In next revision, I will add '-ldl' into tools/testing/selftests/kvm/Makefile.


> > @@ -36,6 +37,8 @@ static __thread volatile struct rseq __rseq = {
> >    */
> >   #define NR_TASK_MIGRATIONS 100000
> >   
> > +static bool __rseq_ownership;
> > +static volatile struct rseq *__rseq_info;
> >   static pthread_t migration_thread;
> >   static cpu_set_t possible_mask;
> >   static int min_cpu, max_cpu;
> > @@ -49,11 +52,33 @@ static void guest_code(void)
> >   		GUEST_SYNC(0);
> >   }
> >   
> > +static void sys_rseq_ownership(void)
> > +{
> > +	long *offset;
> > +	unsigned int *size, *flags;
> > +
> > +	offset = dlsym(RTLD_NEXT, "__rseq_offset");
> > +	size = dlsym(RTLD_NEXT, "__rseq_size");
> > +	flags = dlsym(RTLD_NEXT, "__rseq_flags");
> > +
> > +	if (offset && size && *size && flags) {
> > +		__rseq_ownership = false;
> > +		__rseq_info = (struct rseq *)((uintptr_t)__builtin_thread_pointer() +
> > +					      *offset);
>
> __builtin_thread_pointer doesn't work on all architectures/GCC versions.
> Is this a problem for selftests?

It's a problem as the test case is running on all architectures. I think I
need introduce our own __builtin_thread_pointer() for where it's not
supported: (1) PowerPC  (2) x86 without GCC 11

Please let me know if I still have missed cases where __buitin_thread_pointer()
isn't supported?

Thanks,
Gavin