On 8/19/20 1:52 PM, Luc Van Oostenryck wrote:
> On Wed, Aug 19, 2020 at 10:06:55PM +0200, Luc Van Oostenryck wrote:
>> On Wed, Aug 19, 2020 at 12:15:46PM -0700, Randy Dunlap wrote:
>>> On Linux kernel tree v5.9-rc1, with sparse v0.6.2-180-g49f7e13a,
>>> I see this sparse warning which I don't grok:
>>>
>>> ../include/uapi/asm-generic/signal-defs.h:19:29: CK: error: multiple address spaces given
>>>
>>> for this source code:
>>>
>>> typedef void __signalfn_t(int);
>>> typedef __signalfn_t __user *__sighandler_t; <<<<< line 19
>>>
>>> Are there multiple address spaces there? What are they?
>>>
>>>
>>>
>>> or: is the warning related to the other nearby warnings? (e.g.:)
>>>
>>> ../kernel/signal.c:541:53: CK: warning: incorrect type in initializer (different address spaces)
>>> ../kernel/signal.c:541:53: CK: expected struct k_sigaction *ka
>>> ../kernel/signal.c:541:53: CK: got struct k_sigaction [noderef] __rcu *
>>> ../include/uapi/asm-generic/signal-defs.h:19:29: CK: error: multiple address spaces given
>>> ../kernel/signal.c:694:33: CK: warning: incorrect type in argument 1 (different address spaces)
>>> ../kernel/signal.c:694:33: CK: expected struct spinlock [usertype] *lock
>>> ../kernel/signal.c:694:33: CK: got struct spinlock [noderef] __rcu *
>>
>> I would guess that this __sighandler_t is used somewhere in a
>> struct which is itself accessed via a __rcu pointer.
>> But that would not explain the level of dereference.
>>
>> I'll need to check this one.
>
> The location of the warning should be the assignment in kernel/signal.c:69
>
> static void __user *sig_handler(struct task_struct *t, int sig)
> {
> return t->sighand->action[sig - 1].sa.sa_handler;
> }
>
> There is a lot of type abuses in this file, between 'void __user *' and
> __sighandler_t, but the root cause of the warning is the '__rcu' in
>
> struct task_struct {
> ...
> struct sighand_struct __rcu *sighand;
> ...
> }
OK, thanks for all of your excavation work. :)
--
~Randy
