On Wed, Aug 24, 2022 at 2:49 PM Christophe Leroy
<christophe.leroy@xxxxxxxxxx> wrote:
>
>
>
> Le 24/08/2022 à 08:41, Zhouyi Zhou a écrit :
> > Thank Christophe for reviewing my patch
> >
> > On Wed, Aug 24, 2022 at 2:10 PM Christophe Leroy
> > <christophe.leroy@xxxxxxxxxx> wrote:
> >>
> >>
> >>
> >> Le 24/08/2022 à 07:56, Zhouyi Zhou a écrit :
> >>> In function walk_hugetlb_range, the local variable err may
> >>> be used uninitialzed when:
> >>> ops->pte_hole in side of "else if (ops->pte_hole)" is false.
> >>>
> >>> Signed-off-by: Zhouyi Zhou <zhouzhouyi@xxxxxxxxx>
> >>> ---
> >>> Dear mm Developers:
> >>>
> >>> When I build kernel using "make CC=clang-14"
> >>> the compiler complains following:
> >>>
> >>> CC mm/pagewalk.o
> >>> mm/pagewalk.c:318:12: error: variable 'err' is used uninitialized whenever 'if' condition is false [-Werror,-Wsometimes-uninitialized]
> >>> else if (ops->pte_hole)
> >>> ^~~~~~~~~~~~~
> >>> mm/pagewalk.c:321:7: note: uninitialized use occurs here
> >>> if (err)
> >>> ^~~
> >>> mm/pagewalk.c:318:8: note: remove the 'if' if its condition is always true
> >>> else if (ops->pte_hole)
> >>> ^~~~~~~~~~~~~~~~~~
> >>> mm/pagewalk.c:311:10: note: initialize the variable 'err' to silence this warning
> >>> int err;
> >>> ^
> >>> = 0
> >>> 1 error generated.
> >>> make[1]: *** [scripts/Makefile.build:250: mm/pagewalk.o] Error 1
> >>> make: *** [Makefile:2006: mm] Error 2
> >>>
> >>> I initialize that variable outside of the for loop because we can assign 0 to err
> >>> only once in this function.
> >>>
> >>> After my fix, I can compile the kernel.
> >>>
> >>> Many Thanks
> >>> Zhouyi
> >>> --
> >>> mm/pagewalk.c | 2 +-
> >>> 1 file changed, 1 insertion(+), 1 deletion(-)
> >>>
> >>> diff --git a/mm/pagewalk.c b/mm/pagewalk.c
> >>> index 54b2a1beeeb3..b6eb330e8ecd 100644
> >>> --- a/mm/pagewalk.c
> >>> +++ b/mm/pagewalk.c
> >>> @@ -306,9 +306,9 @@ static int walk_hugetlb_range(unsigned long addr, unsigned long end,
> >>> unsigned long hmask = huge_page_mask(h);
> >>> unsigned long sz = huge_page_size(h);
> >>> const struct mm_walk_ops *ops = walk->ops;
> >>> + int err = 0;
> >>
> >> Why do you move it back outside of the for loop allthough it is
> >> exclusively used inside the loop ?
> > I move it outside of the for loop for performance consideration. Because
> > if we initialize err inside, there will be an assignment statement
> > every iteration.
>
> I think GCC is smart enough to do the assignment only when necessary,
> maybe have a look at the generated assembly in order to confirm.
Thank Christophe for your guidance, I disassembled the GCC generated
code, it optimized away int err = 0 on both X86 and PowerPC platforms.
The disassembled instructions are quite long, so I prefer not to paste
in my email unless someone asks me to do so ;-)
>
> >>
> >>>
> >>> for (; addr < end; addr = next) {
> >>> - int err;
> >>
> >> Another solution would be to add an explicit else, setting err = 0 in
> >> the if/else if sequence.
> > Thank Christophe for your valuable advice, I am going to prepare a 2nd version
>
Thank Christophe for your reminder, I only subscribed to the RCU
mailing list currently, so I submitted duplicated work. I should be
more devoted to the Linux community. I am sorry.
> Don't spend too much time on that, there are already other people
> looking at it, see
> https://lore.kernel.org/linux-mm/20220823153055.2517764-1-nathan@xxxxxxxxxx/T/
>
> >
> > Thanks
> > Zhouyi
Many thanks
Zhouyi
> >>
> >>> pte_t *pte = huge_pte_offset(walk->mm, addr & hmask, sz);
> >>>
> >>> next = hugetlb_entry_end(h, addr, end);
