Generally I2C3 communication can work via I2C3 or I2C4 pins.
Defaults are fine as long as one of the pin groups are floating or grounded or
both share I2C traces.
In NX boards I2C4 SDA is used for GC and connected to 1V8.
So if a GC is slotted, I2C3 works, if not, no communication is possible.
This config was done previously inside I2C3 consumer driver (touchscreen).
Now it's moved inside pinmux_config_i2c.
Apparently, some T210 silicon have an undocumented errata where the MSG_LEFT2/3
registers are not ignored as they should.
When they have random data in POR they cause a hang as long as the message and
SHA calculation speed. So always clear them.
Additionally, clear MSG_LENGTH2/3 registers too even though they do not matter.
Add a 15us worst case scenario delay after OP done for T210.
Practically, because of 1600 MHz RAM, less than 1us delay is needed.
(204 MHz: 15us, 408 MHz: 5us, etc).
- Simplify config and fix partial hashing
- Add partial function calls
- All partial state is handled internally
- All functions now use the classic naming convention.
All operations expect the destination buffer to fit the selected size.
And for simplicity STATE_WAIT is not supported.
Additionally, remove single block ECB and just use the normal function.
The defragmentation runs on every free and it only merges subsequent nodes.
So there's no point of looping all nodes.
So, just check if previous and next nodes are mergeable.
This makes free performant and achieves the same exact functionality.
Utilize charger config set command which allows for supplement mode.
Additionally, remove ping config when SIO and parse buttons/sticks on subcmd reports.
- Add delay after a DSI soft reset
- Remove more duplicated configs
- Improve code to save code size
- Improve comments
- Do not allow display_backlight to be used with OLED
- Convert access from min cluster size to block size
Also allow read/writes to less than cluster size
- In case of intercluster access throw an error
- Do not error on zero size
On reads buffer still needs to be block (instead of cluster) aligned.
On writes, buffer still needs to be readable out of bounds.
