[LTP] [PATCH v2 1/2] syscalls/cma: test case should return TCONF if syscall doesn't exist

[Jan Stancek]

----- Original Message -----
> On Fri, Feb 15, 2019 at 9:53 PM Jan Stancek <jstancek@redhat.com> wrote:
> 
> >
> >
> > ------------------------------
> >
> > Hi Jason,
> >
> > A new failure was occured after applying this new patch.
> >
> > From our observation, the latest process_vm_readv03.c failed with
> > process_vm_readv() Invalid argument error on rhel7.3 platfrom. We suspect
> > that this problem maybe caused by glibc or compiler, because if compile the
> > test program with -O0 it get pass, but with the ltp
> > default optimization -O2 it get fail.  Also, if we revert this new patch,
> > the test PASS whatever with -O2 and -O0, that's very strange.
> >
> >
> > One difference is that the extra function promoted int to long.
> >
> 
> Good catch.
> 
> > Can you try with patch below? (Note that I dropped also root requirement,
> > man page
> > says that it has same requirements as ptrace - you should be able to read
> > from processes you own)
> >
> > This change works for me(one comment in below).
> 
> But there still have a tiny question, in the syscall tracing log, I noticed
> a process_vm_readv() calling does not get this failure in function
> fetch_remote_addrs, what makes it difference with other two calls is
> something I'm thinking now.

"Garbage" on stack most likely. On x86_64, 6th parameter is passed
via stack, glibc copies quadword:

ENTRY (syscall)
        movq %rdi, %rax         /* Syscall number -> rax.  */
        movq %rsi, %rdi         /* shift arg1 - arg5.  */
        movq %rdx, %rsi
        movq %rcx, %rdx
        movq %r8, %r10
        movq %r9, %r8
        movq 8(%rsp),%r9        /* arg6 is on the stack.  */
        syscall                 /* Do the system call.  */

> 
> ====Tracing log without '0L' change====
> 12211 process_vm_readv(12211, 0, 0, 0, 0, 140728898420736) = -1 EINVAL
> (Invalid argument)
> ...
> 12213 process_vm_readv(12212,
> [{"\300`\334\0\0\0\0\0`\2\335\0\0\0\0\0\340\26\335\0\0\0\0\0\220h\335\0\0\0\0\0"...,
> 80}], 1, [{0xdc6060, 80}], 1, 0) = 80
> ...
> 12213 process_vm_readv(12212, 0x7ffecef1e330, 4, 0xdc60c0, 10,
> 140728898420736) = -1 EINVAL (Invalid argument)
> 
> 
> 
> >
> > diff --git a/testcases/kernel/syscalls/cma/process_vm_readv03.c
> > b/testcases/kernel/syscalls/cma/process_vm_readv03.c
> > index 45f7c92e499c..1cec442b3456 100644
> > --- a/testcases/kernel/syscalls/cma/process_vm_readv03.c
> > +++ b/testcases/kernel/syscalls/cma/process_vm_readv03.c
> > @@ -181,7 +181,7 @@ static long *fetch_remote_addrs(void)
> >         remote.iov_len = len;
> >
> >         TEST(ltp_syscall(__NR_process_vm_readv, pids[0], &local,
> > -                        1, &remote, 1, 0));
> > +                        1, &remote, 1, 0L));
> >
> 
> To make things more safely, probably we'd better change all the integer
> parameter too?
> 
> -                        1, &remote, 1, 0));
> +                        1L, &remote, 1L, 0L));

Agreed, let's match all parameters types as before commit:
 9b02cd465f70 ("syscalls/cma: test case should return TCONF if syscall doesn't exist")

> 
> --
> Regards,
> Li Wang
>