Timur Tabi timur@freescale.com wrote on 10/09/2009 18:13:03:
Joakim Tjernlund wrote:
This calculation does not seem to match AN2919.
When I wrote the code, AN2919 was much smaller than what you have today.
Suppose one used only Table 7(almost what we have if you exclude dfsr!= 1) Table 7 is valid for 1 <= dfsr <=5 so how about replacing the current dfsr with: #ifdef __PPC__ u8 dfsr; dfsr = (5*(i2c_clk/1000))/(100000); if (dfsr > 5) dfsr = 5; if (!dfsr) dfsr = 1; debug("i2c_clk:%d, dfsr:%d\n", i2c_clk, dfsr); writeb(dfsr, &dev->dfsrr); /* set default filter */ #endif
The value of FDR is dependent on the value of DFSR, so if I calculate DFSR, I have to also calculate FDR. This means the table goes away. I'm okay with that (since my table is no longer a viable approach, it seems), but it's more work than I'm willing to do at the moment. Especically since this is going to need a lot of testing before I'm willing to push it.
You can manage with the 4 tables listed in the end, they cover all dfsr's, but if you can swing an algorithm that is even better.
Another way of handling this is to edit the table so that it only includes values of DFSR between 1 and 5, which is (unfortunately) *every* entry with a DFSR != 1.
Exactly, that is what I am proposing and that is what the code above does. The entries with DFSR != 1 are likely wrong anyway and is a better fit than todays method.