Re: [PATCH] staging: lustre/lustre/libcfs: Fix type mismatch reported by sparse

["Dilger, Andreas" <andreas.dilger@intel.com>]

On 2015/12/22, 06:05, "Niranjan Dighe" <niranjan.dighe@gmail.com> wrote:

>On Tue, Dec 22, 2015 at 5:14 AM, Greg Kroah-Hartman
><gregkh@linuxfoundation.org> wrote:
>> On Wed, Dec 09, 2015 at 10:38:13PM +0530, Niranjan Dighe wrote:
>>> The third argument to function kportal_memhog_alloc is expected to
>>> be gfp_t whereas the actual argument was unsigned int. Fix this by
>>> explicitly typecasting to gfp_t
>>>
>>> Signed-off-by: Niranjan Dighe <niranjan.dighe@gmail.com>
>>> ---
>>>  drivers/staging/lustre/lustre/libcfs/module.c | 2 +-
>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/staging/lustre/lustre/libcfs/module.c
>>>b/drivers/staging/lustre/lustre/libcfs/module.c
>>> index 96d9d46..9c79f6e 100644
>>> --- a/drivers/staging/lustre/lustre/libcfs/module.c
>>> +++ b/drivers/staging/lustre/lustre/libcfs/module.c
>>> @@ -268,7 +268,7 @@ static int libcfs_ioctl_int(struct cfs_psdev_file
>>>*pfile, unsigned long cmd,
>>>                       /* XXX The ioc_flags is not GFP flags now, need
>>>to be fixed */
>>>                       err = kportal_memhog_alloc(pfile->private_data,
>>>                                                  data->ioc_count,
>>> -                                                data->ioc_flags);
>>> +                                     (__force gfp_t)data->ioc_flags);
>>
>> No, please fix the type to be correct properly, like the comment says
>> needs to be done.
>>
>> thanks,
>>
>> greg k-h
>
>Hello Greg,
>
>I could see that the ioc_flags member of the struct libcfs_ioctl_data
>is used as gfp_t only in the
>case of the ioctl IOC_LIBCFS_MEMHOG. I can think of following ways to
>correct it -
>
>1. Create a union that has 2 different types encapsulated, something like
>this -
>        union {
>                __u32 ioc_flags;
>                gfp_t alloc_flags;
>        }flags;
>Because, the ioc_flags seems to be used in different contexts at
>different places throughout the
>drivers/staging/lustre directory.
>
>2. Is it OK to hardcode the appropriate gfp_t flags for the
>IOC_LIBCFS_MEMHOG, as the userspace
>seems to be taking the decision about the page allocation
>zone/strategy, is this what is intended?

The memhog functionality is used to introduce memory pressure on a client
or server during operation to test error handling as well as memory
allocation deadlocks (e.g. GFP_KERNEL used where GFP_NOFS should be used).
There are other ways to do this in the kernel today, so all of the memhog
code could just be deleted I think.

This looks like kportal_memhog_alloc(), kportal_memhog_free(),
IOC_LIBCFS_MEMHOG, and struct libcfs_device_userstate could be removed.


Cheers, Andreas