On Mon, Oct 24, 2022, at 17:27, Geert Uytterhoeven wrote:
> On Wed, Oct 19, 2022 at 1:17 PM kernel test robot <lkp@xxxxxxxxx> wrote:
>> drivers/net/ethernet/renesas/rswitch.c: In function 'rswitch_ext_desc_get_dptr':
>> >> drivers/net/ethernet/renesas/rswitch.c:355:71: warning: left shift count >= width of type [-Wshift-count-overflow]
>> 355 | return __le32_to_cpu(desc->dptrl) | (dma_addr_t)(desc->dptrh) << 32;
>> | ^~
>> drivers/net/ethernet/renesas/rswitch.c: In function 'rswitch_ext_ts_desc_get_dptr':
>> drivers/net/ethernet/renesas/rswitch.c:367:71: warning: left shift count >= width of type [-Wshift-count-overflow]
>> 367 | return __le32_to_cpu(desc->dptrl) | (dma_addr_t)(desc->dptrh) << 32;
>> | ^~
>>
>>
>> vim +355 drivers/net/ethernet/renesas/rswitch.c
>>
>> 352
>> 353 static dma_addr_t rswitch_ext_desc_get_dptr(struct rswitch_ext_desc *desc)
>> 354 {
>> > 355 return __le32_to_cpu(desc->dptrl) | (dma_addr_t)(desc->dptrh) << 32;
>
> A simple fix would be to replace the cast to "dma_addr_t" by a cast to "u64".
> A more convoluted fix would be:
>
> dma_addr_t dma;
>
> dma = __le32_to_cpu(desc->dptrl);
> if (IS_ENABLED(CONFIG_ARCH_DMA_ADDR_T_64BIT))
> dma |= (u64)desc->dptrh << 32;
> return dma;
>
> Looking at the gcc compiler output, the both cases are optimized to the
> exact same code, for both arm32 and arm64, so I'd go for the simple fix.
>
> BTW, if struct rswitch_ext_desc would just extend struct rswitch_desc,
> you could use rswitch_ext_desc_get_dptr() for both.
>
Regardless of which way this is expressed, it looked like there is
a missing __le32_to_cpu() around the high word.
Arnd
