Re: [PATCH v3] i2c: lpi2c: cache peripheral clock rate

[Alexander Stein <alexander.stein@ew.tq-group.com>]

Hi Marc,

Am Freitag, 21. April 2023, 15:59:52 CEST schrieb Marc Kleine-Budde:
> * PGP Signed by an unknown key
> 
> On 21.04.2023 15:48:59, Alexander Stein wrote:
> [...]
> 
> > Now lpi2c_runtime_resume will call into clk_prepare() which also calls
> > clk_prepare_lock() (identical to clk_get_rate).
> 
> IIRC this is a general problem^w limitation of the clock framework,
> clock providers cannot use clocks themselves in certain callback, e.g.
> set_rate.

Well, that's essentially impossible when this clock provider is attached via 
i2c. i2c transfers potentially need to change or prepare clocks.

One problem is that during clock registration (__clk_core_init) in the call to 
clk_ops.recalc_rate an i2c transfer is issued (via regmap). This is the 
inverse lock order to a regular transfer runtime resuming the i2c master and 
preparing the clocks.

While looking at this, is_prepared is also affected by this. It's reading a 
register, thus issuing a i2c transfer. But this function can also be called 
from within clk_unprepare_unused_subtree, which holds clk_prepare_lock as 
well.

This might be avoided by using regcache, but I'm not really sure this catches 
every case.

Best regards,
Alexander
-- 
TQ-Systems GmbH | Mühlstraße 2, Gut Delling | 82229 Seefeld, Germany
Amtsgericht München, HRB 105018
Geschäftsführer: Detlef Schneider, Rüdiger Stahl, Stefan Schneider
http://www.tq-group.com/