Am 22.03.2014 16:22, schrieb Shuah Khan:
> On 03/22/2014 06:32 AM, Frank Schäfer wrote:
>>
>> Am 21.03.2014 22:04, schrieb Shuah Khan:
>>> Change em28xx_scaler_set() to use em28xx_reg_len() to get register
>>> lengths for EM28XX_R30_HSCALELOW and EM28XX_R32_VSCALELOW registers,
>>> instead of hard-coding the length. Moved em28xx_reg_len() definition
>>> for it to be visible to em28xx_scaler_set().
>>>
>>> Signed-off-by: Shuah Khan <shuah.kh@xxxxxxxxxxx>
>>> ---
>>> drivers/media/usb/em28xx/em28xx-video.c | 29
>>> ++++++++++++++++-------------
>>> 1 file changed, 16 insertions(+), 13 deletions(-)
>>>
>>> diff --git a/drivers/media/usb/em28xx/em28xx-video.c
>>> b/drivers/media/usb/em28xx/em28xx-video.c
>>> index 19af6b3..f8a91de 100644
>>> --- a/drivers/media/usb/em28xx/em28xx-video.c
>>> +++ b/drivers/media/usb/em28xx/em28xx-video.c
>>> @@ -272,6 +272,18 @@ static void em28xx_capture_area_set(struct
>>> em28xx *dev, u8 hstart, u8 vstart,
>>> }
>>> }
>>>
>>> +static int em28xx_reg_len(int reg)
>>> +{
>>> + switch (reg) {
>>> + case EM28XX_R40_AC97LSB:
>>> + case EM28XX_R30_HSCALELOW:
>>> + case EM28XX_R32_VSCALELOW:
>>> + return 2;
>>> + default:
>>> + return 1;
>>> + }
>>> +}
>>> +
>>> static int em28xx_scaler_set(struct em28xx *dev, u16 h, u16 v)
>>> {
>>> u8 mode;
>>> @@ -284,11 +296,13 @@ static int em28xx_scaler_set(struct em28xx
>>> *dev, u16 h, u16 v)
>>>
>>> buf[0] = h;
>>> buf[1] = h >> 8;
>>> - em28xx_write_regs(dev, EM28XX_R30_HSCALELOW, (char *)buf, 2);
>>> + em28xx_write_regs(dev, EM28XX_R30_HSCALELOW, (char *)buf,
>>> + em28xx_reg_len(EM28XX_R30_HSCALELOW));
>>>
>>> buf[0] = v;
>>> buf[1] = v >> 8;
>>> - em28xx_write_regs(dev, EM28XX_R32_VSCALELOW, (char *)buf, 2);
>>> + em28xx_write_regs(dev, EM28XX_R32_VSCALELOW, (char *)buf,
>>> + em28xx_reg_len(EM28XX_R32_VSCALELOW));
>> Hmm... registers 0x30 and 0x32 are always 2 bytes long.
>> So this change would needlessly complicate the code.
>>
>
> The reason I made the change is that em28xx_reg_len() is handling
> these two registers and I thought it would be good to make it
> consistent with other writes to these registers and not hard-code the
> length.
Well, almost all register writes use a hardcoded length (most of the
time it's 1). ;-)
>
> I think it would help with maintenance later by avoiding hard-coding
> the length and use the existing routine that returns the length for
> these registers.
Yes, if the length would be variable, then this would be a good idea.
But AFAIK the length is always 2.
>
> You are correct that it does add a function call in the code path. So
> if you think the trade-off isn't worth it, I am not going to argue
> with it :)
I'm more concerned about the fact that readers of the code could think
that this is a write with a variable length, while the length is
actually always the same.
em28xx_reg_len() is a somewhat dirty hack for vidioc_[g,s]_register
debugging ioctls only.
Btw, what happens when you try to compile the code with this patch
applied and CONFIG_VIDEO_ADV_DEBUG disabled ? ;-)
Anyway, I very much appreciate your efforts to improve the em28xx
driver. :-)
There's much to do !
Regards,
Frank
>
> -- Shuah
>
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
