media: nxp: imx8-isi: fix buiding on 32-bit

  From: Arnd Bergmann <arnd@arndb.de>

The #if check is wrong, leading to a build failure:

drivers/media/platform/nxp/imx8-isi/imx8-isi-hw.c: In function 'mxc_isi_channel_set_inbuf':
drivers/media/platform/nxp/imx8-isi/imx8-isi-hw.c:33:5: error: "CONFIG_ARCH_DMA_ADDR_T_64BIT" is not defined, evaluates to 0 [-Werror=undef]
   33 | #if CONFIG_ARCH_DMA_ADDR_T_64BIT
      |     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~

This could just be an #ifdef, but it seems nicer to just remove
the check entirely. Apparently the only reason for the #ifdef
is to avoid another warning:

drivers/media/platform/nxp/imx8-isi/imx8-isi-hw.c:55:24: error: right shift count >= width of type [-Werror=shift-count-overflow]

But this is best avoided by using the lower_32_bits()/upper_32_bits()
helpers.

Fixes: cf21f328fcaf ("media: nxp: Add i.MX8 ISI driver")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 .../media/platform/nxp/imx8-isi/imx8-isi-hw.c | 41 ++++++++++---------
 1 file changed, 21 insertions(+), 20 deletions(-)
  

Hi Arnd,

Thank you for the patch.

On Tue, Apr 18, 2023 at 12:37:27AM +0200, Arnd Bergmann wrote:
> From: Arnd Bergmann <arnd@arndb.de>
> 
> The #if check is wrong, leading to a build failure:
> 
> drivers/media/platform/nxp/imx8-isi/imx8-isi-hw.c: In function 'mxc_isi_channel_set_inbuf':
> drivers/media/platform/nxp/imx8-isi/imx8-isi-hw.c:33:5: error: "CONFIG_ARCH_DMA_ADDR_T_64BIT" is not defined, evaluates to 0 [-Werror=undef]
>    33 | #if CONFIG_ARCH_DMA_ADDR_T_64BIT
>       |     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
> 
> This could just be an #ifdef, but it seems nicer to just remove
> the check entirely. Apparently the only reason for the #ifdef
> is to avoid another warning:
> 
> drivers/media/platform/nxp/imx8-isi/imx8-isi-hw.c:55:24: error: right shift count >= width of type [-Werror=shift-count-overflow]
> 
> But this is best avoided by using the lower_32_bits()/upper_32_bits()
> helpers.

I've submitted a patch yesterday that uses #ifdef, but I like this one
better. One could argue that it leads to dead code on 32-bit platforms,
but the ISI is only present in 64-bit SoCs, so that's not an issue.

> Fixes: cf21f328fcaf ("media: nxp: Add i.MX8 ISI driver")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
>  .../media/platform/nxp/imx8-isi/imx8-isi-hw.c | 41 ++++++++++---------
>  1 file changed, 21 insertions(+), 20 deletions(-)
> 
> diff --git a/drivers/media/platform/nxp/imx8-isi/imx8-isi-hw.c b/drivers/media/platform/nxp/imx8-isi/imx8-isi-hw.c
> index db538f3d88ec..f6112b83866a 100644
> --- a/drivers/media/platform/nxp/imx8-isi/imx8-isi-hw.c
> +++ b/drivers/media/platform/nxp/imx8-isi/imx8-isi-hw.c
> @@ -29,11 +29,10 @@ static inline void mxc_isi_write(struct mxc_isi_pipe *pipe, u32 reg, u32 val)
>  
>  void mxc_isi_channel_set_inbuf(struct mxc_isi_pipe *pipe, dma_addr_t dma_addr)
>  {
> -	mxc_isi_write(pipe, CHNL_IN_BUF_ADDR, dma_addr);
> -#if CONFIG_ARCH_DMA_ADDR_T_64BIT
> +	mxc_isi_write(pipe, CHNL_IN_BUF_ADDR, lower_32_bits(dma_addr));
>  	if (pipe->isi->pdata->has_36bit_dma)
> -		mxc_isi_write(pipe, CHNL_IN_BUF_XTND_ADDR, dma_addr >> 32);
> -#endif
> +		mxc_isi_write(pipe, CHNL_IN_BUF_XTND_ADDR,
> +			      upper_32_bits(dma_addr));
>  }
>  
>  void mxc_isi_channel_set_outbuf(struct mxc_isi_pipe *pipe,
> @@ -45,34 +44,36 @@ void mxc_isi_channel_set_outbuf(struct mxc_isi_pipe *pipe,
>  	val = mxc_isi_read(pipe, CHNL_OUT_BUF_CTRL);
>  
>  	if (buf_id == MXC_ISI_BUF1) {
> -		mxc_isi_write(pipe, CHNL_OUT_BUF1_ADDR_Y, dma_addrs[0]);
> -		mxc_isi_write(pipe, CHNL_OUT_BUF1_ADDR_U, dma_addrs[1]);
> -		mxc_isi_write(pipe, CHNL_OUT_BUF1_ADDR_V, dma_addrs[2]);
> -#if CONFIG_ARCH_DMA_ADDR_T_64BIT
> +		mxc_isi_write(pipe, CHNL_OUT_BUF1_ADDR_Y,
> +			lower_32_bits(dma_addrs[0]));

Could you please align this with the ( ? I can also do so in my tree,
but I expect Mauro to pick this patch directly from the list, so a v2
would make it easier. You can then add my

Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>

> +		mxc_isi_write(pipe, CHNL_OUT_BUF1_ADDR_U,
> +			lower_32_bits(dma_addrs[1]));
> +		mxc_isi_write(pipe, CHNL_OUT_BUF1_ADDR_V,
> +			lower_32_bits(dma_addrs[2]));
>  		if (pipe->isi->pdata->has_36bit_dma) {
>  			mxc_isi_write(pipe, CHNL_Y_BUF1_XTND_ADDR,
> -				      dma_addrs[0] >> 32);
> +				      upper_32_bits(dma_addrs[0]));
>  			mxc_isi_write(pipe, CHNL_U_BUF1_XTND_ADDR,
> -				      dma_addrs[1] >> 32);
> +				      upper_32_bits(dma_addrs[1]));
>  			mxc_isi_write(pipe, CHNL_V_BUF1_XTND_ADDR,
> -				      dma_addrs[2] >> 32);
> +				      upper_32_bits(dma_addrs[2]));
>  		}
> -#endif
>  		val ^= CHNL_OUT_BUF_CTRL_LOAD_BUF1_ADDR;
>  	} else  {
> -		mxc_isi_write(pipe, CHNL_OUT_BUF2_ADDR_Y, dma_addrs[0]);
> -		mxc_isi_write(pipe, CHNL_OUT_BUF2_ADDR_U, dma_addrs[1]);
> -		mxc_isi_write(pipe, CHNL_OUT_BUF2_ADDR_V, dma_addrs[2]);
> -#if CONFIG_ARCH_DMA_ADDR_T_64BIT
> +		mxc_isi_write(pipe, CHNL_OUT_BUF2_ADDR_Y,
> +			      lower_32_bits(dma_addrs[0]));
> +		mxc_isi_write(pipe, CHNL_OUT_BUF2_ADDR_U,
> +			      lower_32_bits(dma_addrs[1]));
> +		mxc_isi_write(pipe, CHNL_OUT_BUF2_ADDR_V,
> +			      lower_32_bits(dma_addrs[2]));
>  		if (pipe->isi->pdata->has_36bit_dma) {
>  			mxc_isi_write(pipe, CHNL_Y_BUF2_XTND_ADDR,
> -				      dma_addrs[0] >> 32);
> +				      upper_32_bits(dma_addrs[0]));
>  			mxc_isi_write(pipe, CHNL_U_BUF2_XTND_ADDR,
> -				      dma_addrs[1] >> 32);
> +				      upper_32_bits(dma_addrs[1]));
>  			mxc_isi_write(pipe, CHNL_V_BUF2_XTND_ADDR,
> -				      dma_addrs[2] >> 32);
> +				      upper_32_bits(dma_addrs[2]));
>  		}
> -#endif
>  		val ^= CHNL_OUT_BUF_CTRL_LOAD_BUF2_ADDR;
>  	}
>
  

On Tue, Apr 18, 2023, at 05:20, Laurent Pinchart wrote:
> On Tue, Apr 18, 2023 at 12:37:27AM +0200, Arnd Bergmann wrote:
>> drivers/media/platform/nxp/imx8-isi/imx8-isi-hw.c:55:24: error: right shift count >= width of type [-Werror=shift-count-overflow]
>> 
>> But this is best avoided by using the lower_32_bits()/upper_32_bits()
>> helpers.
>
> I've submitted a patch yesterday that uses #ifdef, but I like this one
> better. One could argue that it leads to dead code on 32-bit platforms,
> but the ISI is only present in 64-bit SoCs, so that's not an issue.

I think there are actually some people using 32-bit kernels on imx8,
but they tend to do it for the wrong reasons, and I'm not worried
about wasting a few bytes in those setups.

My first draft kept an IS_ENABLED(CONFIG_ARCH_DMA_ADDR_T_64BIT) inside
each of the three if(), but I dropped that for readability.

>>  	if (buf_id == MXC_ISI_BUF1) {
>> -		mxc_isi_write(pipe, CHNL_OUT_BUF1_ADDR_Y, dma_addrs[0]);
>> -		mxc_isi_write(pipe, CHNL_OUT_BUF1_ADDR_U, dma_addrs[1]);
>> -		mxc_isi_write(pipe, CHNL_OUT_BUF1_ADDR_V, dma_addrs[2]);
>> -#if CONFIG_ARCH_DMA_ADDR_T_64BIT
>> +		mxc_isi_write(pipe, CHNL_OUT_BUF1_ADDR_Y,
>> +			lower_32_bits(dma_addrs[0]));
>
> Could you please align this with the ( ? I can also do so in my tree,
> but I expect Mauro to pick this patch directly from the list, so a v2
> would make it easier. You can then add my
>
> Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>

Ok, I sent that out now, but now I see I forgot to add your
Reviewed-by.

     Arnd