Re: [PATCH v2] fbdev: bfin-lq035q1-fb: new Blackfin Landscape LCD EZ-Extender driver

[Mike Frysinger <vapier.adi@gmail.com>]

On Thu, Sep 24, 2009 at 19:32, Andrew Morton wrote:
> On Thu, 17 Sep 2009 17:37:06 -0400 Mike Frysinger wrote:
>> +config FB_BFIN_LQ035Q1
>> +     tristate "SHARP LQ035Q1DH02 TFT LCD"
>> +     depends on FB && BLACKFIN
>> +     select FB_CFB_FILLRECT
>> +     select FB_CFB_COPYAREA
>> +     select FB_CFB_IMAGEBLIT
>> +     select BFIN_GPTIMERS
>> +     select SPI
>
> Are we sure about the `select SPI'?  There's only one other place in
> the kernel which does this, and `select' often makes things explode.  I
> fear that you're either selecting the wrong thing or you're selecting
> something which won't work well.

is it there on purpose and is it not just a mistaken typo ?  yes.  do
we really need it and will we cry if it changes to a "depends" ?  no.

it's just confusing to have a device driver disappear if SPI is
disabled, but considering SPI is enabled by default now, it's not a
big deal.

>> +#define DRIVER_NAME "bfin-lq035q1"
>> +static char driver_name[] = DRIVER_NAME;
>
> Will the compielr magically put this string into read-only storage for
> us, or should we do that manually with `const'?

is this question a rhetorical one ?  oh no, infinite loop ...

i'll fix it up in v3

>> +static int lq035q1_control(unsigned char reg, unsigned short value)
>> +{
>> +     int ret;
>> +     u8 regs[3] = {LQ035_INDEX, 0, 0};
>> +     u8 dat[3] = {LQ035_DATA, 0, 0};
>> +
>> +     if (spi_control.spidev) {
>> +             regs[2] = reg;
>> +             dat[1] = value >> 8;
>> +             dat[2] = value & 0xFF;
>> +
>> +             ret = spi_write(spi_control.spidev, regs, ARRAY_SIZE(regs));
>> +             ret |= spi_write(spi_control.spidev, dat, ARRAY_SIZE(dat));
>> +     } else
>> +             return -ENODEV;
>> +
>> +     return ret;
>> +}
>
> I am suspecting that this function (and the similar ones below) rely
> upon state within the hardware and will hence misbehave if two
> instances are run concurrently.
>
> Is that correct>  If so, is there locking to prevent this from occurring?

if by "instances" you mean "users" as in "multiple programs
reading/writing the framebuffer concurrently", then probably.  rather
than handle the locking ourselves, it can be pushed to the SPI bus by
having the regs/dat be transfers in a single message.
-mike