Re: [viro-vfs:misc.cmpxchg 4/8] arch/sparc/include/asm/cmpxchg_32.h:51:61: sparse: sparse: cast truncates bits from constant value (eb9f becomes 9f)

[Al Viro <viro@zeniv.linux.org.uk>]

On Wed, Apr 03, 2024 at 03:40:22AM +0800, kernel test robot wrote:
> tree:   https://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git misc.cmpxchg
> head:   9e2f22ef1ae21b949a3903727d7e7cd5eb48810f
> commit: 1b2857f7277164d8bed4e831e3c3696572a51f0b [4/8] sparc32: add __cmpxchg_u{8,16}() and teach __cmpxchg() to handle those sizes
> config: sparc-randconfig-r132-20240403 (https://download.01.org/0day-ci/archive/20240403/202404030332.d8MKrNbM-lkp@intel.com/config)
> compiler: sparc-linux-gcc (GCC) 13.2.0
> reproduce: (https://download.01.org/0day-ci/archive/20240403/202404030332.d8MKrNbM-lkp@intel.com/reproduce)


> sparse warnings: (new ones prefixed by >>)
>    kernel/bpf/helpers.c:2157:18: sparse: sparse: symbol 'bpf_task_release_dtor' was not declared. Should it be static?
>    kernel/bpf/helpers.c:2187:18: sparse: sparse: symbol 'bpf_cgroup_release_dtor' was not declared. Should it be static?
>    kernel/bpf/helpers.c: note: in included file (through arch/sparc/include/asm/cmpxchg.h, arch/sparc/include/asm/atomic_32.h, arch/sparc/include/asm/atomic.h, ...):
> >> arch/sparc/include/asm/cmpxchg_32.h:51:61: sparse: sparse: cast truncates bits from constant value (eb9f becomes 9f)
>    kernel/bpf/helpers.c: note: in included file (through include/linux/timer.h, include/linux/workqueue.h, include/linux/bpf.h):
>    include/linux/list.h:83:21: sparse: sparse: self-comparison always evaluates to true
>    include/linux/list.h:83:21: sparse: sparse: self-comparison always evaluates to true
>    kernel/bpf/helpers.c: note: in included file (through arch/sparc/include/asm/cmpxchg.h, arch/sparc/include/asm/atomic_32.h, arch/sparc/include/asm/atomic.h, ...):
> >> arch/sparc/include/asm/cmpxchg_32.h:51:61: sparse: sparse: cast truncates bits from constant value (eb9f becomes 9f)
>    kernel/bpf/helpers.c:2495:18: sparse: sparse: context imbalance in 'bpf_rcu_read_lock' - wrong count at exit
>    kernel/bpf/helpers.c:2500:18: sparse: sparse: context imbalance in 'bpf_rcu_read_unlock' - unexpected unlock
> 
> vim +51 arch/sparc/include/asm/cmpxchg_32.h
> 
>     44	
>     45	/* don't worry...optimizer will get rid of most of this */
>     46	static inline unsigned long
>     47	__cmpxchg(volatile void *ptr, unsigned long old, unsigned long new_, int size)
>     48	{
>     49		switch (size) {
>     50		case 1:
>   > 51			return __cmpxchg_u8((u8 *)ptr, (u8)old, (u8)new_);
>     52		case 2:
>     53			return __cmpxchg_u16((u16 *)ptr, (u16)old, (u16)new_);
>     54		case 4:
>     55			return __cmpxchg_u32((u32 *)ptr, (u32)old, (u32)new_);
>     56		default:
>     57			__cmpxchg_called_with_bad_pointer();
>     58			break;
>     59		}
>     60		return old;
>     61	}

... and calls are cmpxchg(&node->owner, NULL, BPF_PTR_POISON).  IOW, that's
a false positive - size will be 4, not 1.

IMO it's a sparse bug; one way to work around that would've been

        return
                size == 1 ? __cmpxchg_u8(ptr, old, new_) :
                size == 2 ? __cmpxchg_u16(ptr, old, new_) :
                size == 4 ? __cmpxchg_u32(ptr, old, new_) :
                        (__cmpxchg_called_with_bad_pointer(), old);

instead of that switch - elimination of dead subexpressions *does* suppress
such warnings.  Elimination of dead branches doesn't...