On Fri, May 03, 2019 at 07:38:52AM +0300, Kalle Valo wrote:
> Nick Desaulniers <ndesaulniers@xxxxxxxxxx> writes:
>
> > On Thu, May 2, 2019 at 8:16 AM Nathan Chancellor
> > <natechancellor@xxxxxxxxx> wrote:
> >>
> >> When building with -Wuninitialized, Clang warns:
> >>
> >> drivers/net/wireless/rsi/rsi_91x_sdio.c:940:43: warning: variable 'data'
> >> is uninitialized when used here [-Wuninitialized]
> >> put_unaligned_le32(TA_HOLD_THREAD_VALUE, data);
> >> ^~~~
> >> drivers/net/wireless/rsi/rsi_91x_sdio.c:930:10: note: initialize the
> >> variable 'data' to silence this warning
> >> u8 *data;
> >> ^
> >> = NULL
> >> 1 warning generated.
> >>
> >> Using Clang's suggestion of initializing data to NULL wouldn't work out
> >> because data will be dereferenced by put_unaligned_le32. Use kzalloc to
> >> properly initialize data, which matches a couple of other places in this
> >> driver.
> >>
> >> Fixes: e5a1ecc97e5f ("rsi: add firmware loading for 9116 device")
> >> Link: https://github.com/ClangBuiltLinux/linux/issues/464
> >> Signed-off-by: Nathan Chancellor <natechancellor@xxxxxxxxx>
> >> ---
> >> drivers/net/wireless/rsi/rsi_91x_sdio.c | 21 ++++++++++++++-------
> >> 1 file changed, 14 insertions(+), 7 deletions(-)
> >>
> >> diff --git a/drivers/net/wireless/rsi/rsi_91x_sdio.c b/drivers/net/wireless/rsi/rsi_91x_sdio.c
> >> index f9c67ed473d1..b35728564c7b 100644
> >> --- a/drivers/net/wireless/rsi/rsi_91x_sdio.c
> >> +++ b/drivers/net/wireless/rsi/rsi_91x_sdio.c
> >> @@ -929,11 +929,15 @@ static int rsi_sdio_ta_reset(struct rsi_hw *adapter)
> >> u32 addr;
> >> u8 *data;
> >>
> >> + data = kzalloc(sizeof(u32), GFP_KERNEL);
> >
> > Something fishy is going on here. We allocate 4 B but declare data as
> > a u8* (pointer to individual bytes)? In general, dynamically
> > allocating that few bytes is a code smell; either you meant to just
> > use the stack, or this memory's lifetime extends past the lifetime of
> > this stackframe, at which point you probably just meant to stack
> > allocate space in a higher parent frame and pass this preallocated
> > memory down to the child frame to get filled in.
> >
> > Reading through this code, I don't think that the memory is meant to
> > outlive the stack frame. Is there a reason why we can't just declare
> > data as:
> >
> > u8 data [4];
> >
> > then use ARRAY_SIZE(data) or RSI_9116_REG_SIZE in rsi_reset_chip(),
> > getting rid of the kzalloc/kfree?
>
> I haven't checked the details but AFAIK stack variables are not supposed
> to be used with DMA. So in that case I think it's ok alloc four bytes,
> unless the DMA rules have changed of course. But I didn't check if rsi
> is using DMA here, just a general comment.
>
> --
> Kalle Valo
I don't think it is using the DMA API but it might be the same thing for
SDIO. If passing that around on the stack is okay, great but we don't
want what commit f700546682a6 ("rsi: fix nommu_map_sg overflow kernel
panic") fixes to happen here.
I can't answer that for sure though since I am not at all familiar with
this driver or the SDIO APIs.
Cheers,
Nathan
