[LTP] {get|set}_thread_area issue on ARM

[Yury Norov]

On Fri, Dec 04, 2015 at 08:19:55AM -0500, Jan Stancek wrote:
> 
> 
> ----- Original Message -----
> > From: "Yury Norov" <ynorov@caviumnetworks.com>
> > To: "Jan Stancek" <jstancek@redhat.com>
> > Cc: ltp@lists.linux.it, "David Howells" <dhowells@redhat.com>
> > Sent: Friday, 4 December, 2015 2:02:41 PM
> > Subject: Re: [LTP] {get|set}_thread_area issue on ARM
> > 
> > On Fri, Dec 04, 2015 at 04:20:37AM -0500, Jan Stancek wrote:
> > > 
> > > 
> > > 
> > > 
> > > ----- Original Message -----
> > > > From: "Yury Norov" <ynorov@caviumnetworks.com>
> > > > To: ltp@lists.linux.it
> > > > Cc: "David Howells" <dhowells@redhat.com>
> > > > Sent: Friday, 4 December, 2015 4:47:11 AM
> > > > Subject: [LTP] {get|set}_thread_area issue on ARM
> > > > 
> > > > Hi all,
> > > > 
> > > > I just tried build ltp for arm, and one (maybe, first of) test failed
> > > > build. Namely, set_thread_area01.c
> > > > 
> > > > The reason is that __NR_set_thread_area and __NR_get_thread_area are
> > > > not declared. They should come from
> > > > /usr/arm-linux-gnueabi/include/asm/unistd.h
> > > > But there's only:
> > > > [...]
> > > > #define __NR_remap_file_pages           (__NR_SYSCALL_BASE+253)
> > > >                                          /* for set_thread_area */
> > > >                                          /* for get_thread_area */
> > > > #define __NR_set_tid_address            (__NR_SYSCALL_BASE+256)
> > > > [...]
> > > > 
> > > > In kernel, there are stubs for corresponding syscall numbers. So it
> > > > looks like this pair is deprecated and unsupported. I can build ltp
> > > > for aarch64 and x84_64 just fine.
> > > 
> > > Looking at history, it appears it was never supported.
> > > 
> > > > 
> > > > Could someone explain me what am I doing wrong?
> > > 
> > > Can you try to add:
> > >   #include "linux_syscall_numbers.h"
> > > to testcases/kernel/syscalls/set_thread_area/set_thread_area01.c
> > > and see if it compiles. And when you run it, it should end with TCONF.
> > > 
> > > Regards,
> > > Jan
> > 
> > Hi Jan,
> > 
> > It doesn't help much. The problem is not in that ltp doesn't know
> > symbols, but because they are identical - both are -1.
> >   (__LTP__NR_INVALID_SYSCALL)
> 
> Hi,
> 
> That was expected.
> 
> > 
> > And gcc treats is as an error:
> > arm-linux-gnueabi-gcc -g -O2 -g -O2 -fno-strict-aliasing -pipe -Wall
> > -W -g -O2 -Wold-style-definition
> > -I/home/yury/work/ltp/testcases/kernel/include -I../../../../include
> > -I../../../../include -Wl,--rpath=/root/lib32
> > -Wl,--dynamic-linker=/root/lib32/ld-2.21.so  -L../../../../lib
> > set_thread_area01.c   -lltp -o set_thread_area01
> > set_thread_area01.c: In function ‘get_name’:
> > set_thread_area01.c:70:2: error: duplicate case value
> >   case __NR_get_thread_area:
> >   ^
> > set_thread_area01.c:67:2: error: previously used here
> >   case __NR_set_thread_area:
> >   ^
> > set_thread_area01.c: In function ‘main’:
> > set_thread_area01.c:87:17: warning: comparison between signed
> > and unsigned integer expressions [-Wsign-compare]
> >    for (i = 0; i < sizeof(tests) / sizeof(struct test); i++) {
> >                  ^
> > <builtin>: recipe for target 'set_thread_area01' failed
> 
> This wasn't - I didn't realize they are used in switch.
> 
> I suggest to remove the get_name() function and store names
> in struct test (with some macro to stringify those syscall __NR_#).
> 
> Would you care to send a patch?
> 

This weekend, I think.

> > 
> > BTW, I forgot to mention, if I "uncomment" that definitions in system
> > headers, everything becomes fine.
> 
> I assume you still get TCONF when you run it, because kernel should
> return ENOSYS for both.
> 
> Regards,
> Jan
> 
> > 
> > Preprocessed sources for arm and x86_64 are attached.
> > 
> > BR,
> > Yury.
> > 
> > 
> >