Bluetooth: Controller: Fix HCI command buffer allocation failure

[cvinayak]

Fix HCI command buffer allocation failure, that can cause loss of Host Number of Completed Packets command.

Fail by rejecting the HCI Host Buffer Size command if the required number of HCI command buffers are not allocated in the Controller implementation.

Relates to commit 8161430 ("Bluetooth: Add workaround
for no command buffer available")'.

Relates to commit 297f4f4 ("Bluetooth: Split HCI
command & event buffers to two pools").

Relates to #81866

One of many CI failure exposing incorrect allocation:

tests/bsim/bluetooth/host/l2cap/einprogress/test_scripts/run.sh FAILED (0.838 s)
d_01: @00:00:00.000000  *** Booting Zephyr OS build v4.0.0-3133-ge9f6c00b05bc ***
d_01: @00:00:00.000000 INFO: Test start: tester
d_00: @00:00:00.000000  *** Booting Zephyr OS build v4.0.0-3133-ge9f6c00b05bc ***
d_00: @00:00:00.000000 INFO: Test start: dut
d_01: @00:00:00.005768  [00:00:00.005,767] <wrn> [sysworkq] net_buf: Timeout discarded. No blocking in syswq
d_01: @00:00:00.005768  [00:00:00.005,767] <wrn> [sysworkq] net_buf: Timeout discarded. No blocking in syswq
d_01: @00:00:00.005768  [00:00:00.005,767] <wrn> [sysworkq] net_buf: Timeout discarded. No blocking in syswq
d_01: @00:00:00.005768  [00:00:00.005,767] <wrn> [sysworkq] net_buf: Timeout discarded. No blocking in syswq
d_01: @00:00:00.005768  [00:00:00.005,767] <err> [sysworkq] bt_ctlr_hci: FC: Require Host ACL packets (2) < CONFIG_BT_BUF_CMD_TX_COUNT (2)
d_01: @00:00:00.005768  [00:00:00.005,767] <wrn> [main] bt_hci_core: opcode 0x0c33 status 0x12 
d_01: @00:00:00.005768  ASSERTION FAIL [!err] @ WEST_TOPDIR/zephyr/tests/bsim/bluetooth/host/l2cap/einprogress/src/tester.c:51
d_01: @00:00:00.005768  @ WEST_TOPDIR/zephyr/lib/os/assert.c:43
d_01: @00:00:00.005768  [00:00:00.005,767] <err> [main] os: >>> ZEPHYR FATAL ERROR 4: Kernel panic on CPU 0
d_01: @00:00:00.005768  [00:00:00.005,767] <err> [main] os: Current thread: 0x826bc40 (main)
d_01: @00:00:00.005768  [00:00:00.005,767] <err> [main] os: Halting system
d_00: @00:00:00.005768  [00:00:00.005,767] <wrn> [sysworkq] net_buf: Timeout discarded. No blocking in syswq
d_00: @00:00:00.005768  [00:00:00.005,767] <wrn> [sysworkq] net_buf: Timeout discarded. No blocking in syswq
d_00: @00:00:00.005768  [00:00:00.005,767] <wrn> [sysworkq] net_buf: Timeout discarded. No blocking in syswq
d_00: @00:00:00.005768  [00:00:00.005,767] <wrn> [sysworkq] net_buf: Timeout discarded. No blocking in syswq
d_00: @00:00:00.005768  [00:00:00.005,767] <err> [sysworkq] bt_ctlr_hci: FC: Require Host ACL packets (2) < CONFIG_BT_BUF_CMD_TX_COUNT (2)
d_00: @00:00:00.005768  [00:00:00.005,767] <wrn> [main] bt_hci_core: opcode 0x0c33 status 0x12 
d_00: @00:00:00.005768  ASSERTION FAIL [!err] @ WEST_TOPDIR/zephyr/tests/bsim/bluetooth/host/l2cap/einprogress/src/dut.c:108
d_00: @00:00:00.005768  @ WEST_TOPDIR/zephyr/lib/os/assert.c:43
d_00: @00:00:00.005768  [00:00:00.005,767] <err> [main] os: >>> ZEPHYR FATAL ERROR 4: Kernel panic on CPU 0
d_00: @00:00:00.005768  [00:00:00.005,767] <err> [main] os: Current thread: 0x826bc40 (main)
d_00: @00:00:00.005768  [00:00:00.005,767] <err> [main] os: Halting system
p_2G4: WARNING: (src/bs_pc_base.c:437): Device 0 left the party unsuspectingly.. I treat it as if it disconnected
timeout: the monitored command dumped core
/__w/zephyr/zephyr/tests/bsim/sh_common.source: line 22: 413863 Trace/breakpoint trap   $@
p_2G4: WARNING: (src/bs_pc_base.c:437): Device 1 left the party unsuspectingly.. I treat it as if it disconnected
timeout: the monitored command dumped core
/__w/zephyr/zephyr/tests/bsim/sh_common.source: line 22: 413864 Trace/breakpoint trap   $@

[jhedberg]

This paragraph isn't completely clear to me. Even if a controller supports ncmd > 1, the Zephyr host has a command queue and will always round down the value to 1, or rather, treat it as a boolean:

zephyr/subsys/bluetooth/host/hci_core.c

Lines 2499 to 2503 in ddd795a

	/* Allow next command to be sent */
	if (ncmd) {
	k_sem_give(&bt_dev.ncmd_sem);
	bt_tx_irq_raise();
	}

The other help text paragraphs sort of make sense, but I just wanted to make sure the above host implementation detail is clear, so that this PR isn't making any false assumptions (since I haven't quite wrapped my head around all the details of it).

[cvinayak]

Yes, Zephyr Host only implements ncmd = 1. See,

zephyr/subsys/bluetooth/host/hci_core.c

Lines 4340 to 4348 in ddd795a

	/* Give cmd_sem allowing to send first HCI_Reset cmd, the only
	* exception is if the controller requests to wait for an
	* initial Command Complete for NOP.
	*/
	if (!IS_ENABLED(CONFIG_BT_WAIT_NOP)) {
	k_sem_init(&bt_dev.ncmd_sem, 1, 1);
	} else {
	k_sem_init(&bt_dev.ncmd_sem, 0, 1);
	}

I guess it is not difficult to use the same semaphore and we able to support ncmd > 1, but we do not have support in native Controller today for that though.

[cvinayak]

the Zephyr host has a command queue and will always round down the value to 1, or rather, treat it as a boolean:

@jhedberg I believe the implementation "can" support ncmd > 1, but today ncmd in the evt is only used as a bool to pause the queuing, but the ncmd_sem if initialized correctly would support ncmd > 1. But we are not going towards that in this PR.

[Thalley]

Why can't we use BT_BUF_CMD_TX_COUNT instead?

[cvinayak]

I should rename this to BT_CTLR_HCI_NUM_CMD_PKT_MAX, right? This is the supported 1 normal HCI command flow control count.

BT_BUF_CMD_TX_COUNT will be the buffer counts calculated with/without flow control support.

[Thalley]

A promptless Kconfig with a static value doesn't really have any value (i.e. nonconfigurable) - It could just as well be a #define or something.

[cvinayak]

As the derived value BT_BUF_CMD_TX_COUNT in hci_common_internal.h is needed by hci_raw.c (and a Controller implementation HCI driver glue) which will be used with other vendor Controllers (source or binary) to build Controller only builds (hci_ipc, hci_uart, etc...), I do not want to hard code using #define which will require hci_raw.c to include numerous vendor Controller's header file. Kconfig is cleaner to fetch a Controller's implementation specifics (just like we have the numerous _SUPPORT Kconfigs).