On Mon, 14 Dec 2020 12:15:55 +0300, Serge Semin wrote: > Current clocks description doesn't provide a comprehensive notion about > what "stmmaceth" and "pclk" actually represent from the IP-core manual > point of view. The bindings file states: > stmmaceth - "GMAC main clock", > apb - "Peripheral registers interface clock". > It isn't that easy to understand what they actually mean especially seeing > the DW *MAC manual operates with clock definitions like Application, > System, Host, CSR, Transmit, Receive, etc clocks. Moreover the clocks > usage in the driver doesn't shade a full light on their essence. What > inferred from there is that the "stmmaceth" name has been assigned to the > common clock, which feeds both system and CSR interfaces. But what about > "apb"? The bindings defines it as the clock for "peripheral registers > interface". So it's close to the CSR clock in the IP-core manual notation. > If so then when "apb" clock is specified aside with the "stmmaceth", it > represents a case when the DW *MAC is synthesized with CSR_SLV_CLK=y > (separate system and CSR clocks). But even though the "apb" clock is > requested in the MAC driver, the driver doesn't actually use it as a > separate CSR clock where the IP-core manual requires. All of that makes me > thinking that the case of separate system and CSR clocks isn't correctly > implemented in the driver. > > Let's start with elaborating the clocks description so anyone reading > the DW *MAC bindings file would understand that "stmmaceth" is the > system clock and "pclk" is actually the CSR clock. Indeed in accordance > with sheets depicted in [1]: > system/application clock can be either of: hclk_i, aclk_i, clk_app_i; > CSR clock can be either of: hclk_i, aclk_i, clk_app_i, clk_csr_i. > (Most likely the similar definitions present in the others IP-core > manuals.) So the CSR clock can be tied to the application clock > considering the later as the main clock, but not the other way around. In > case if there is only "stmmaceth" clock specified in a DT node, then it > will be considered as a source of clocks for both application and CSR. But > if "pclk" is also specified in the list of the device clocks, then it will > be perceived as the separate CSR clock. > > [1] DesignWare Cores Ethernet MAC Universal Databook, Revision 3.73a, > October 2013, p. 564. > > Signed-off-by: Serge Semin <Sergey.Semin@xxxxxxxxxxxxxxxxxxxx> > --- > .../devicetree/bindings/net/snps,dwmac.yaml | 12 ++++++++++-- > 1 file changed, 10 insertions(+), 2 deletions(-) > Reviewed-by: Rob Herring <robh@xxxxxxxxxx>
