Hi, Arnd

On 2016/5/11 22:50, Arnd Bergmann wrote:

On Wednesday 11 May 2016 19:16:44 Zhangjian wrote:

quoted

Hi,

On 2016/5/11 18:12, Zhangjian (Bamvor) wrote:

quoted

Hi, Arnd

On 2016/5/11 16:09, Arnd Bergmann wrote:
  > On Wednesday 11 May 2016 10:04:16 Zhangjian wrote:
  >>> I don't remember. It's probably not important whether we have the shift
  >>> in there, as long as it's independent of the actual kernel page size and
  >>> user space and kernel agree on the calling conventions.
  >> Well. I am ok with where to shift the pages size because we get the same
  >> result. I was just thinking if we should get rid of the name of mmap2 in our
  >> ILP32 porting. Actually, it is mmap but we name it as mmap2. User may confused
  >> if they do not know the implementations.
  >
  > That is a good point: If the implementation matches the mmap() behavior rather than
  > mmap2(), we should rename the macro by doing
  >
  > #undef __NR_mmap2
  > #define __NR_mmap 222
  >
  > in the uapi/asm/unistd.h file for ilp32 mode.
Do you mean define the following things in kernel:

diff --git a/arch/arm64/include/uapi/asm/unistd.h b/arch/arm64/include/uapi/asm/unistd.h
index 1caadc2..3f79640 100644
--- a/arch/arm64/include/uapi/asm/unistd.h
+++ b/arch/arm64/include/uapi/asm/unistd.h
@@ -14,3 +14,9 @@
    * along with this program.  If not, see <http://www.gnu.org/licenses/>.
    */
   #include <asm-generic/unistd.h>
+
+#ifdef __ILP32__
+#undef __NR_mmap2
+#define __NR_mmap 222
+#endif /* #ifdef __ILP32__ */
+

Then glibc could call mmap instead of mmap2.
I could not try it now. Because after change off_t to 64bit in glibc, stat
is fail. I may need to revert the stat relative patch.

After revert stat relative patch in glibc, mmap01-mmap14 success. But mmap16
success with segfault. I will investigate it later.

There is pointer and size_t in mmap, so, IIUC, we need to clear the top halves
of register by using COMPAT_SYSCALL_WRAP6.

Correct, good catch!

quoted

And after check the function in
arch/s390/kernel/compat_linux.c, I feel that we need to do the same thing for
pread64 and pwrite64.

quoted

But I got following error when I try to add
COMPAT_SYSCALL_WRAP4(pread64, unsigned int, fd, char __user *, buf,
size_t, count, loff_t, pos);
COMPAT_SYSCALL_WRAP4(pwrite64, unsigned int, fd, const char __user *, buf,
size_t, count, loff_t, pos);

Hmm, that is indeed tricky. I think COMPAT_SYSCALL_WRAP4 rightfully
refuses the loff_t argument here, as the common case is that this is
not possible.

It works if I apply the following patch, I defined the wrong `__TYPE_IS_xxx`
yesterday. Should we merge this into ILP32 series or send the compat.h
and syscalls.h individually? The current series of ILP32 is a little bit
long and hard to review.

diff --git a/include/linux/compat.h b/include/linux/compat.h
index ba6ebe0..22a9565 100644
--- a/include/linux/compat.h
+++ b/include/linux/compat.h

@@ -747,7 +747,8 @@ asmlinkage long compat_sys_fanotify_mark(int, unsigned int, __u32, __u32,
  #ifndef __SC_COMPAT_CAST
  #define __SC_COMPAT_CAST(t, a) ({                                      \
         BUILD_BUG_ON((sizeof(t) > 4) && !__TYPE_IS_L(t) &&              \
-                    !__TYPE_IS_UL(t) && !__TYPE_IS_PTR(t));            \
+                    !__TYPE_IS_UL(t) && !__TYPE_IS_PTR(t) &&           \
+                    !__TYPE_IS_LOFFT(t));                              \
         ((t) ((t)(-1) < 0 ? (s64)(s32)(a) : (u64)(u32)(a)));            \
  })
  #endif

diff --git a/include/linux/syscalls.h b/include/linux/syscalls.h
index 6e57d9c..66eb85d 100644
--- a/include/linux/syscalls.h
+++ b/include/linux/syscalls.h

@@ -47,6 +47,7 @@
  #define __TYPE_IS_L(t) (__same_type((t)0, 0L))
  #define __TYPE_IS_UL(t)        (__same_type((t)0, 0UL))
  #define __TYPE_IS_LL(t) (__same_type((t)0, 0LL) || __same_type((t)0, 0ULL))
+#define __TYPE_IS_LOFFT(t)     (__same_type((t)0, (loff_t)0))
  #define __SC_LONG(t, a) __typeof(__builtin_choose_expr(__TYPE_IS_LL(t), 0LL, 0L)) a
  #define __SC_CAST(t, a)        (t) a
  #define __SC_ARGS(t, a)        a

diff --git a/kernel/compat_wrapper.c b/kernel/compat_wrapper.c
index 98b68b8..28f02d0 100644
--- a/kernel/compat_wrapper.c
+++ b/kernel/compat_wrapper.c

@@ -304,3 +304,7 @@ COMPAT_SYSCALL_WRAP3(getpeername, int, fd, struct sockaddr __user *, usockaddr,
  COMPAT_SYSCALL_WRAP6(sendto, int, fd, void __user *, buff, size_t, len,
                      unsigned int, flags, struct sockaddr __user *, addr,
                      int, addr_len);
+COMPAT_SYSCALL_WRAP4(pread64, unsigned int, fd, char __user *, buf,
+                       size_t, count, loff_t, pos);
+COMPAT_SYSCALL_WRAP4(pwrite64, unsigned int, fd, const char __user *, buf,
+                        size_t, count, loff_t, pos);
 >
 > Can you open-code this using a COMPAT_SYSCALL4 definition similar to what
 > arch/tile has, but without the merging of the two halves of the argument?

I am lost here. Tile do not use the wrapper, and it do not use the loff_t
either:
COMPAT_SYSCALL_DEFINE6(pread64, unsigned int, fd, char __user *, ubuf,
                        size_t, count, u32, dummy, u32, low, u32, high)

Regards

Bamvor

	Arnd