mirror of
https://github.com/brain-hackers/u-boot-brain
synced 2024-09-13 00:03:26 +09:00
beb90a53f3
For the RK3399, i2c_set_rate (and by extension: our spi_set_rate,
which had been mindlessly following the template of the i2c_set_rate
implementation) miscalculates the rate returned due to a off-by-one
error resulting from the following sequence of events:
1. calculates 'src_div := src_freq / target_freq'
2. stores 'src_div - 1' into the register (the actual divider applied
in hardware is biased by adding 1)
3. returns the result of the DIV_RATE(src_freq, src_div) macro, which
expects the (decremented) divider from the hardware-register and
implictly adds 1 (i.e. 'DIV_RATE(freq, div) := freq / (div + 1)')
This can be observed with the SPI driver, which sets a rate of 99MHz
based on the GPLL frequency of 594MHz: the hardware generates a clock
of 99MHz (src_div is 6, the bitfield in the register correctly reads 5),
but reports a frequency of 84MHz (594 / 7) on return.
To fix, we have two options:
* either we bias (i.e. "DIV_RATE(GPLL, src_div - 1)"), which doesn't
make for a particularily nice read
* we simply call the i2c/spi_get_rate function (introducing additional
overhead for the additional register-read), which reads the divider
from the register and then passes it through the DIV_RATE macro
Given that this code is not time-critical, the more readable solution
(i.e. calling the appropriate get_rate function) is implemented in this
change.
Signed-off-by: Philipp Tomsich <philipp.tomsich@theobroma-systems.com>
Tested-by: Klaus Goger <klaus.goger@theobroma-systems.com>
Acked-by: Simon Glass <sjg@chromium.org>
|
||
|---|---|---|
| .. | ||
| clk_rk3036.c | ||
| clk_rk3188.c | ||
| clk_rk3288.c | ||
| clk_rk3328.c | ||
| clk_rk3399.c | ||
| Makefile | ||