nimble/ll: Add support for custom TX power configuration

[sjanc]

This allows to provide custom TX power setting and rounding functions which can take into account FEM configuration. Since this is highly dependend on HW configuration (FEM type and mode, BPFs, PA power etc) there is no point in providing "default" implementation as this should be typically provided by BSP or application.

[apache-mynewt-bot]

Style check summary

Our coding style is here!

nimble/controller/src/ble_ll_dtm.c

@@ -34,16 +34,16 @@
 #include "ble_ll_priv.h"
 
 STATS_SECT_START(ble_ll_dtm_stats)
-    STATS_SECT_ENTRY(rx_count)
-    STATS_SECT_ENTRY(tx_failed)
-    STATS_SECT_ENTRY(rx_failed)
+STATS_SECT_ENTRY(rx_count)
+STATS_SECT_ENTRY(tx_failed)
+STATS_SECT_ENTRY(rx_failed)
 STATS_SECT_END
 STATS_SECT_DECL(ble_ll_dtm_stats) ble_ll_dtm_stats;
 
 STATS_NAME_START(ble_ll_dtm_stats)
-    STATS_NAME(ble_ll_dtm_stats, rx_count)
-    STATS_NAME(ble_ll_dtm_stats, tx_failed)
-    STATS_NAME(ble_ll_dtm_stats, rx_failed)
+STATS_NAME(ble_ll_dtm_stats, rx_count)
+STATS_NAME(ble_ll_dtm_stats, tx_failed)
+STATS_NAME(ble_ll_dtm_stats, rx_failed)
 STATS_NAME_END(ble_phy_stats)
 
 struct dtm_ctx {

[andrzej-kaczmarek]

so I thought it over and I think it should work a bit different:

• LL should always set TX power instead of relying on FEM to do this
• we should have API to get PA/LNA gain, LL should call it and then calculate actual TX power or RSSI using that value and also RF path compensation, if set by host
• optionally we can have syscfg to define constant gain for PA/LNA to have some build time optimizations possible

proposed names for APIs would be ble_ll_fem_(pa|lna)_gain_get()
syscfg BLE_LL_FEM_(PA|LNA)_GAIN - if set to 0 means LL has to call API to get actual gain, otherwise use that value

[sjanc]

The thing is that FEM might not have fixed gain, ie in SKY66112 TX gain depends on SKY configuration and input power, so LL will have to iterate (?) to match expected output power with whatever can be set by phy, alternatively we can add some output parameters where FEM returns also expected PHY tx power

[andrzej-kaczmarek]

The "current" solution will call FEM API which will set tx power via phy. Proposed solution will call FEM API to obtain tx power and set it via phy - I suspect this will be the same value, so what is the problem here?

[sjanc]

"LL should call it and then calculate actual TX power or RSSI using that value and also RF path compensation, if set by host"

this is other way around, only FEM can calculate actual TX power since it (may) depends on input power and FEM (possibly runtime) configuration.

Also to cover all scenarios, like rounding, reading supported power ranges etc we need ble_ll_fem_pa_gain_set() which will configure FEM as needed (as oppose to get() which would just return values for rounding)

Than, ble_ll_fem_pa_gain_get() would return actually configured gain (compared to PHY tx power that was last "set")

[sjanc]

this will be handled in #1374