Kmod driver at iicbus. attach() and config_intrhook(9)

[Alexander Mishin]

Ian Lepore писал 2020-09-03 18:48:
> On Thu, 2020-09-03 at 15:02 +0400, Alexander Mishin wrote:
>> Andriy Gapon писал 2020-09-03 11:29:
>> > On 02/09/2020 01:18, John-Mark Gurney wrote:
>> > > So my reading of the twsi driver is that it handles things correctly
>> > > allowing devices pre-interrupts to attach, and using interrupts for
>> > > transfers post-cold.
>> >
>> > I just want to note that twsi uses _completely_ different code for
>> > polled and
>> > interrupt modes.  So, while either mode should be active at correct
>> > times, I
>> > cannot certify that the polled mode works correctly for all transfer
>> > types.
>> > I haven't reviewed the polled mode code.
>> 
>> This adds some more light on my question.
>> 
>> As a result:
>> For twsi I use iicbb_transfer() (as Emmanuel Vadot recommended to
>> someone on the mailing list).
>> It is definitely not works for my driver, until interrupts went 
>> enabled.
>> But it works perfectly well with the config_intrhook().
>> 
>> I have never tried iicbus_write() at boot time yet, but I think I'll
>> try, just to know for the future.
>> 
>> Thanks
>> 
> 
> iicbb_transfer() is for bit-bang i2c; hopefully that was a typo.
> 
> All i2c slave drivers should be using iicbus_transfer_excl() or the
> iicdev_readfrom()/iicdev_writeto() helper functions.  The only time to
> use iicbus_transfer() directly is when you need to conduct a dialog
> with one or more slaves using multiple transactions and it's important
> to keep the bus locked across the entire series of transactions for
> some reason (meaning you must handle the acquire/release bus stuff
> yourself).  There is virtually never a reason to call the low-level
> start/stop/read/write functions directly.
> 
> It's a pity that this stuff is so horribly documented.  It has evolved
> extensively over the years, not always in good ways, and documentation
> and examples either don't exist or haven't kept up.
> 
> Bottom line: if the twsi driver has the config_intrhook logic in it
> then your driver shouldn't need it.  If you do need it, something is
> broken, either in your driver or in the twsi driver, and we should
> figure out what the real problem is.
> 
> -- Ian

Not a typo :-(

Many, many thanks for such a portion of valuable information.
Went to look for examples in the kernel sources.