Re: [PATCH 08/10] Use __kernel_ulong_t in struct msqid64_ds

From: H.J. Lu
Date: Thu May 17 2012 - 23:43:10 EST


On Thu, May 17, 2012 at 8:39 PM, H.J. Lu <hjl.tools@xxxxxxxxx> wrote:
> On Thu, May 17, 2012 at 8:21 PM, H. Peter Anvin <hpa@xxxxxxxxxx> wrote:
>> On 05/17/2012 04:51 PM, H. Peter Anvin wrote:
>>>
>>> This patch and the one before it seems to have another problem: we
>>> currently define __BITS_PER_LONG as:
>>>
>>> #ifdef __x86_64__
>>> # define __BITS_PER_LONG 64
>>> #else
>>> # define __BITS_PER_LONG 32
>>> #endif
>>>
>>
>> H.J., do you see any problem *other* than this wretched struct
>> msqid64_ds with changing the above from __x86_64__ to
>>
>> #if defined(__x86_64__) && !defined(__ILP32__)
>>
>> ... in the above?
>>
>> As far as struct msqid64_ds,  I think we can fix it simply because x86
>> is the only compat-aware architecture which has to deal with it.
>>
>> (Incidentally, if sh is ever expanded to 64 bits, it will have a problem
>> in the bigendian configuration...)
>
> That will be wrong.   __BITS_PER_LONG defines # bits of long
> as seen by kernel.  We don't use it in user space.  Remember
> x32 uses the identical interface as x86-64.  So
>
> #ifdef __x86_64__
> # define __BITS_PER_LONG 64
> #else
> # define __BITS_PER_LONG 32
> #endif
>
> struct msqid64_ds {
>        struct ipc64_perm msg_perm;
>        __kernel_time_t msg_stime;      /* last msgsnd time */
> #if __BITS_PER_LONG != 64
>        unsigned long   __unused1;
> #endif
>        __kernel_time_t msg_rtime;      /* last msgrcv time */
> #if __BITS_PER_LONG != 64
>        unsigned long   __unused2;
> #endif
>        __kernel_time_t msg_ctime;      /* last change time */
> #if __BITS_PER_LONG != 64
>        unsigned long   __unused3;
> #endif
>
> are absolutely correct for x32.  You can think
>
> #if __BITS_PER_LONG != 64
>
> as
>
> #ifndef __x86_64__
>
> which is used in glibc.
>

Now I remembered. Here is another patch. Here

#if __BITS_PER_LONG == 64

means

#ifdef __x86_64__

Change __BITS_PER_LONG to __BITS_PER_LONG_AS_SEEN_BY_KERNEL
may avoid this confusion.

--
H.J.
---
commit eadbf4f44bd12ec379f43f9359849a5012403c50
Author: H.J. Lu <hjl.tools@xxxxxxxxx>
Date: Fri Apr 27 09:58:59 2012 -0700

Define __statfs_word as __kernel_long_t

When __BITS_PER_LONG is 64, define __statfs_word as __kernel_long_t
instead of long for user space.

Signed-off-by: H.J. Lu <hjl.tools@xxxxxxxxx>

diff --git a/include/asm-generic/statfs.h b/include/asm-generic/statfs.h
index c749af9..2043d51 100644
--- a/include/asm-generic/statfs.h
+++ b/include/asm-generic/statfs.h
@@ -16,7 +16,7 @@ typedef __kernel_fsid_t fsid_t;
*/
#ifndef __statfs_word
#if __BITS_PER_LONG == 64
-#define __statfs_word long
+#define __statfs_word __kernel_long_t
#else
#define __statfs_word __u32
#endif
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/