On Mon, Dec 18, 2017 at 10:57:55PM +0100, Wolfram Sang wrote: > The current calculation for I2C bus speeds does not match the datasheet and is, > in deed, a little too fast for 100kHz settings (~107kHz have been measured). > > This series implements a second calculation according to the datasheet. In > order to not cause regressions on the various old SoCs, using this new formula > is opt-in per SoC. Once the new formula has been verified to work, we can > switch over. By default, the old behaviour is retained. > > So, the first patches are refactoring the driver to support per-SoC frequency > calculations. This is done by making the already existing per-SoC > setup()-callback mandatory and let the frequency calculation be done in there. > > This series has been tested on a R-Car H2 Lager board (r8a7790). The computed > values match the suggested values in the datasheet. Also, measurements showed > that the bus frequency is now 100kHz. For 400 kHz, the overall bus speed did > not change much, but the low-to-high relationship is now matching the > datasheet, too. > > This series is RFC because I think we need to discuss when to enable a SoC to > the new formula. For Lager, I did some basic checking but did not run a full > V4L testsuite or other complex configurations making use of I2C clients. Gen3, > in theory, should be easier to test because only the PMIC and an EEPROM are > connected to IIC_DVFS. However, there are no testpoints available for measuring > on Salvator-X(S). And the suggested values in the datasheet do not match the > module clock we have. Do ULCB boards maybe have testpoints for IIC_DVFS? For > Lager, it could be argued that the resulting speed is slower than before which > should not introduce regressions. But I am open to discussion here. > > A branch can be found here: > > git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux.git renesas/iic-refactor > > Thanks, and looking forward to comments > > Wolfram > > > Wolfram Sang (7): > i2c: sh_mobile: move type detection upwards > i2c: sh_mobile: allow setup callback to return errno > i2c: sh_mobile: require setup callback > i2c: sh_mobile: let RuntimePM do the clock handling > i2c: sh_mobile: add helper to check frequency calculations > i2c: sh_mobile: add new frequency calculation for later SoC > i2c: sh_mobile: let r8a7790 (R-Car H2) use the new formula > Fixed a spelling mistake in the patch description of patch 5 and applied to for-next.
Attachment:
signature.asc
Description: PGP signature
