Re: [PATCH net-next] net: can: Fix compiling warning

From: Kees Cook
Date: Mon Aug 12 2019 - 13:19:31 EST


On Wed, Aug 07, 2019 at 01:50:42PM +0300, Dan Carpenter wrote:
> On Tue, Aug 06, 2019 at 06:41:44PM +0200, Oliver Hartkopp wrote:
> > I compiled the code (the original version), but I do not get that "Should it
> > be static?" warning:
> >
> > user@box:~/net-next$ make C=1
> > CALL scripts/checksyscalls.sh
> > CALL scripts/atomic/check-atomics.sh
> > DESCEND objtool
> > CHK include/generated/compile.h
> > CHECK net/can/af_can.c
> > ./include/linux/sched.h:609:43: error: bad integer constant expression
> > ./include/linux/sched.h:609:73: error: invalid named zero-width bitfield
> > `value'
> > ./include/linux/sched.h:610:43: error: bad integer constant expression
> > ./include/linux/sched.h:610:67: error: invalid named zero-width bitfield
> > `bucket_id'
> > CC [M] net/can/af_can.o
>
> The sched.h errors suppress Sparse warnings so it's broken/useless now.
> The code looks like this:
>
> include/linux/sched.h
> 613 struct uclamp_se {
> 614 unsigned int value : bits_per(SCHED_CAPACITY_SCALE);
> 615 unsigned int bucket_id : bits_per(UCLAMP_BUCKETS);
> 616 unsigned int active : 1;
> 617 unsigned int user_defined : 1;
> 618 };
>
> bits_per() is zero and Sparse doesn't like zero sized bitfields.

I just noticed these sparse warnings too -- what's happening here? Are
they _supposed_ to be 0-width fields? It doesn't look like it to me:

CONFIG_UCLAMP_BUCKETS_COUNT=5
...
#define UCLAMP_BUCKETS CONFIG_UCLAMP_BUCKETS_COUNT
...
unsigned int bucket_id : bits_per(UCLAMP_BUCKETS);

I would expect this to be 3 bits wide. ... Looks like gcc agrees:

struct uclamp_se {
unsigned int value:11; /* 0: 0 4 */
unsigned int bucket_id:3; /* 0:11 4 */
...

So this is a sparse issue?

--
Kees Cook