Le 15/06/2021 à 09:21, Christoph Hellwig a écrit :

On Tue, Jun 15, 2021 at 09:03:42AM +0200, Christophe Leroy wrote:

quoted

Le 15/06/2021 ?? 08:52, Christoph Hellwig a ??crit??:

quoted

On Tue, Jun 15, 2021 at 06:41:01AM +0000, Christophe Leroy wrote:

quoted

+	unsafe_copy_to_user(__ucs_to, __ucs_from,			\
+			    sizeof(struct kernel_siginfo), label);	\
+	unsafe_clear_user(__ucs_expansion, SI_EXPANSION_SIZE, label);	\
+} while (0)

unsafe_clear_user does not exist at this point, and even your later
patch only adds it for powerpc.

You missed below chunck I guess:

quoted

diff --git a/include/linux/uaccess.h b/include/linux/uaccess.h
index c05e903cef02..37073caac474 100644
--- a/include/linux/uaccess.h
+++ b/include/linux/uaccess.h

@@ -398,6 +398,7 @@ long strnlen_user_nofault(const void __user *unsafe_addr, long count);
   #define unsafe_put_user(x,p,e) unsafe_op_wrap(__put_user(x,p),e)
   #define unsafe_copy_to_user(d,s,l,e) unsafe_op_wrap(__copy_to_user(d,s,l),e)
   #define unsafe_copy_from_user(d,s,l,e) unsafe_op_wrap(__copy_from_user(d,s,l),e)
+#define unsafe_clear_user(d, l, e) unsafe_op_wrap(__clear_user(d, l), e)

That doesn't help with architectures that define user_access_begin but
do not define unsafe_clear_user. (i.e. x86).

Yes, the day they want to use unsafe_copy_siginfo_to_user() they'll have to implement 
unsafe_clear_user().

Until that day, they don't need unsafe_clear_user() and I'm sure the result would be disastrous if a 
poor powerpc guy like me was trying to implement some low level x86 code.

Similar to unsafe_get_compat_sigset(), an arch wanting to use it has to implement 
unsafe_copy_from_user().