I received a bug report for an unrelated patch when used with m68k-nommu.
It appears that the origin of the problem is that __get_user() and
__put_user() doesn't handle correctly __user. These 2 patches fix this.
Note: this is only minimaly tested but is quite straightforward and
since this only change __user annotation it will not change the
generated code.
Changes since v1:
* fix typo: s/plan/plain/
* appease checkpatch with better style: s/__force*/__force */
* avoid excessive line length caused by the added cast.
Luc Van Oostenryck (2):
m68k,nommu: add missing __user in uaccess' __ptr() macro
m68k,nommu: fix implicit cast from __user in __{get,put}_user_asm()
arch/m68k/include/asm/uaccess_no.h | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
--
2.26.2
The assembly for __get_user() & __put_user() uses a macro, __ptr(),
to cast the pointer to 'unsigned long *' but the pointer is always
a __user one and so this cast creates a lot of warnings when using
Sparse.
So, change to the cast to 'unsigned long __user *'.
Reported-by: kbuild test robot <[email protected]>
Signed-off-by: Luc Van Oostenryck <[email protected]>
---
arch/m68k/include/asm/uaccess_no.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/arch/m68k/include/asm/uaccess_no.h b/arch/m68k/include/asm/uaccess_no.h
index a24cfe4a0d32..9651766a62af 100644
--- a/arch/m68k/include/asm/uaccess_no.h
+++ b/arch/m68k/include/asm/uaccess_no.h
@@ -60,7 +60,7 @@ extern int __put_user_bad(void);
* aliasing issues.
*/
-#define __ptr(x) ((unsigned long *)(x))
+#define __ptr(x) ((unsigned long __user *)(x))
#define __put_user_asm(err,x,ptr,bwl) \
__asm__ ("move" #bwl " %0,%1" \
--
2.26.2
The assembly for __get_user_asm() & __put_user_asm() uses memcpy()
when the size is 8.
However, the pointer is always a __user one while memcpy() expects
a plain one and so this cast creates a lot of warnings when using
Sparse.
So, fix this by adding a cast to 'void __force *' at memcpy()'s
argument.
Reported-by: kbuild test robot <[email protected]>
Signed-off-by: Luc Van Oostenryck <[email protected]>
---
arch/m68k/include/asm/uaccess_no.h | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/arch/m68k/include/asm/uaccess_no.h b/arch/m68k/include/asm/uaccess_no.h
index 9651766a62af..9959327e99b0 100644
--- a/arch/m68k/include/asm/uaccess_no.h
+++ b/arch/m68k/include/asm/uaccess_no.h
@@ -42,7 +42,7 @@ static inline int _access_ok(unsigned long addr, unsigned long size)
__put_user_asm(__pu_err, __pu_val, ptr, l); \
break; \
case 8: \
- memcpy(ptr, &__pu_val, sizeof (*(ptr))); \
+ memcpy((void __force *)ptr, &__pu_val, sizeof(*(ptr))); \
break; \
default: \
__pu_err = __put_user_bad(); \
@@ -85,7 +85,7 @@ extern int __put_user_bad(void);
u64 l; \
__typeof__(*(ptr)) t; \
} __gu_val; \
- memcpy(&__gu_val.l, ptr, sizeof(__gu_val.l)); \
+ memcpy(&__gu_val.l, (const void __force *)ptr, sizeof (__gu_val.l)); \
(x) = __gu_val.t; \
break; \
} \
--
2.26.2
Hi Luc,
On 30/5/20 5:02 am, Luc Van Oostenryck wrote:
> I received a bug report for an unrelated patch when used with m68k-nommu.
> It appears that the origin of the problem is that __get_user() and
> __put_user() doesn't handle correctly __user. These 2 patches fix this.
>
> Note: this is only minimaly tested but is quite straightforward and
> since this only change __user annotation it will not change the
> generated code.
>
>
> Changes since v1:
> * fix typo: s/plan/plain/
> * appease checkpatch with better style: s/__force*/__force */
> * avoid excessive line length caused by the added cast.
>
> Luc Van Oostenryck (2):
> m68k,nommu: add missing __user in uaccess' __ptr() macro
> m68k,nommu: fix implicit cast from __user in __{get,put}_user_asm()
>
> arch/m68k/include/asm/uaccess_no.h | 6 +++---
> 1 file changed, 3 insertions(+), 3 deletions(-)
Looks good, thanks for the fixup.
Applied to the m68knommu git tree, for-next branch.
Regards
Greg