Bluetooth: controller: split: Fix central assert on multiple reconnections

[cvinayak]

This reverts commit 7417e6e.
This reverts commit d3e3f8d.

Updated the lll_conn_flush interface to pass the connection
handle while the LLL connection context stored handle has
been invalidated.

Fixes #22003.

Signed-off-by: Vinayak Kariappa Chettimada [email protected]

[alexandru-porosanu-nxp]

I've tested on VEGABoard and the previous behavior (the assertion) doesn't happen. I guess that Nordic is fixed as well.

[wopu-ot]

I to not think that the PR solves the issue the original PR tried to solve: the mayfly_enqueue in ull_conn.c:1640 may be chained instead of being called inline, which causes the LLL to see the already invalidated handle. As far as I can tell, ll_conn_handle_get(conn) in line 1742 could have the same problem and some other mechanism is needed to ensure that lll_conn_flush sees the handle intact.

[George-Stefan]

ll_conn_handle_get(conn) returns the correct handle even if lll->handle is invalid.
In lll_conn_flush(), use the new handle param, not lll->handle

[wopu-ot]

ll_conn_handle_get(conn) returns the correct handle even if lll->handle is invalid.
In lll_conn_flush(), use the new handle param, not lll->handle

I stand corrected. I will need some time to think through the interaction with our LLL.

[bu5hm4nn]

👍 This pull request fixes crashes we were having after a few minutes of continuous transfer when running HCI_UART on nRF52810 with BlueZ 5.52 and connecting with mutliple devices especially Android devices.

[mtpr-ot]

This PR also fixes race condition where slave ull_conn_done is invoked after flush. This would give an assert in ticker_update_conn_op_cb, as it attempted to update the ticker drift after ticker node had been stopped. I strongly recommend to merge this PR - @wopu-ot, any objections?