Now that the incoming ACL & HCI event pools can be defined externally
to hci_core.c there should be convenience macros to know what's the
minimum required buffer size for these pools.
Change-Id: I6f2a7322a3d77e1304e9d925767a1fd3471c76c7
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
So far the assumption has been that the host stack manages all
incoming and outgoing buffers. For the incoming buffers (from the
controller) this has required hci_core.c to manage its own pools and
do the host flow control. This setup makes perfect sense for an
architecture where the controller resides remotely on a different CPU
& address space (i.e. the "traditional" HCI transport case).
When the stack runs on a system where the controller resides in the
same address space this setup doesn't work that well. In such a
scenario the incoming buffers are ideally created as low down in the
stack as possible (i.e. below HCI), which means that the current
hci_core.c cannot be responsible for managing their pools.
To allow for both types of architectures this patch introduces a new
BLUETOOTH_HOST_BUFFERS Kconfig option that can be selected to say that
host-side management is desired, or deselected to say that the
controller (residing in the same address space) takes care of managing
the incoming buffers.
So far the incoming buffer types were identified by hci_core.c by
looking at their "free pool" pointers, however as soon as the pools
are allowed to be somewhere else this doesn't work. To solve this we
now require a minimum user data size for all Bluetooth buffers and use
that to store the buffer type.
Change-Id: I14bc32007e3e3f17c654f71f79b520650028d7ce
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
There's no need to maintain a separate module for this anymore.
Subsequent patches will have further simplifications thanks to this
being now handled in a single c-file.
Change-Id: I4510c9f72b121e1e5fd19eeb8b5d5ddf2f4bfffe
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
To avoid unnecessary stack variables and to simplify code, add helper
macros to access the Bluetooth specific ACL and HCI user data that's
part of the buffers.
Change-Id: Idc337537721115c4091982c04f8a05bbbf0d5a19
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Now that all code is converted remove the temporary helpers for the
old bt_buf API.
Change-Id: I5183add527d204c51103549fda3378040d5f8f68
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
This patch performs the minimal necessary conversion to the net_buf
API. It uses a temporary "#define bt_buf net_buf" to make it possible
to convert the code in smaller increments. Most old bt_buf function
also serve as one-line wrappers to the matching net_buf APIs. Once
everything is converted these helpers will be removed.
Change-Id: Ie31433d33576022c9c193a35d2389267005545d6
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Change all the Intel and Wind River code license from BSD-3 to Apache 2.
Change-Id: Id8be2c1c161a06ea8a0b9f38e17660e11dbb384b
Signed-off-by: Javier B Perez Hernandez <javier.b.perez.hernandez@linux.intel.com>
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Signed-off-by: Allan Stephens <allan.stephens@windriver.com>
Signed-off-by: Benjamin Walsh <benjamin.walsh@windriver.com>
This adds bt_buf_clone which can be used to duplicate a buffer, including
its contents.
Change-Id: I473841e1fce00fddbfac1fb2c55d40741658484f
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Enables directly add 16-bit value at the end of operational buffer.
Value is converted to little-endian format before is appended.
Position of buffer's tail is updated during addition.
Usage of bt_buf_add() call enables monitoring available limited room
in the buffer when CONFIG_BLUETOOTH_DEBUG is on.
Change-Id: Id7ac62cd3b11ef03302e94185a7faae747eccb20
Signed-off-by: Arkadiusz Lichwa <arkadiusz.lichwa@tieto.com>
Change all occurances of /*! to /** to match javadoc
style.
Change-Id: I3a759d34e0e928216f61252682266e64c5b875f8
Signed-off-by: Anas Nashif <anas.nashif@intel.com>
Reorganize and compress some of the buffer variable to achieve smaller
struct size.
Change-Id: I1ce711c22844470403b085b0f13e77f762f25262
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
The public header file buf.h uses stdint.h definitions and should
therefore have the necessary include statement (since public header
files are supposed to be self-contained).
Change-Id: I9e2a9e13e5d2c35aa61c7552c3ba9fa73b8886a6
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
To make the bt_buf APIs consistent with the recent bt_but_add changes
and more user friendly, change the pull() and push() return values
also to 'void *'.
Change-Id: I25242635c87882db99152735c22316d4de00d363
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Almost all callers of bt_buf_add end up having to do explicit type
casts of the return value since 'uint8_t *' rarely fits what's needed.
Changing the return type to 'void *' removes this extra type casting
need and makes the API friendlier to its users.
Change-Id: I6a79605006153b6d09f0d662a051cb599af68c37
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
bt_buf_pull_le16 convert the initial 16 bits to little endian.
Change-Id: I6f6e9d63fc19b0579d9f445e15e975929fc63ba9
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Instead of having global command completion handler we can keep the
code more concise by letting bt_hci_cmd_send_sync() give back the
return parameters straight to the caller. This patch extends the
buf->hci.sync variable to not just contain the initial semaphore, but
to have the return parameter bt_buf once the semaphore is given back.
The bt_hci_cmd_send_sync() function receives an extra parameter which
may be NULL if the caller isn't interested in the return parameters.
Change-Id: I41e3e4b1189c4a19d0159947a03fd7052cc98fa9
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
In certain scenarios we want to keep the buffers around for longer
and avoid the default bt_buf_put() calls from putting them back to the
available buffers queue. This patch adds reference counting for the
buffers, along with a bt_buf_hold() API to increment the reference
count. Now bt_buf_put() will only put the buffer back to the pool if
the reference count hits 0.
Change-Id: I1590c5574e18600939f55e5339a6da3d061f7682
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
We're already sending 'dummy' buffers to wake up the connection
fibers. The problem is that if we provide a real buffer type it may
have unintended consequences. E.g. with BT_ACL_OUT when bt_buf_put()
is called it may trigger an invalid HCI_Host_Num_Completed_Packets
command.
This patch creates a new BT_DUMMY buffer type to avoid this scenario.
The new type is an alias for BT_CMD since the pool where these buffers
come from is not quota-controlled and synced up with the controller.
Change-Id: I42ac235811dc242345853f7858c10bfc43e4f07d
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
To track which ACL handle each buffer belongs to (needed for host flow
control) add the necessary information into a new ACL-specific struct.
Change-Id: Ie6cc7c32a70b43a4ff5954bb9dca34e4f62da292
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
To accommodate for ACL user data move the HCI command/event data into
its own struct and put it inside a union in bt_buf.
Change-Id: I680500b15709d14b1e9f70ced88664d607a6568c
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
In order to do proper flow control of ACL data to/from the controller
we need to have precise management of the available buffers. Mixing
with the events/commands buffers would make this impossible. This
patch splits the buffer pools into three separate ones:
1. HCI commands/events
2. Incoming ACL data
3. Outgoing ACL data
The total number of available buffers is also increased to match
what's the smallest number supported by current controllers (to avoid
the stack from becoming a bottle neck).
Change-Id: I7e131d61c83a4dda554068d7917c5ee09f2f837d
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
We may soon want to have a _wait() variant of bt_buf_get, so to avoid
the number of 'get' function growing too large consolidate the
existing get() and get_reserve() functions into a single one. The new
consolidated function also takes the type as input parameter so that
we know this from the very start and thereby plan for the split into
multiple buffer pools.
Change-Id: Ia09448565349def2be9bc08d9510fedd029480b4
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
We'll soon want to have dedicated pools for outgoing and incoming ACL
data. To know from which pool to get and put the buffers each buffer
should contain enough information to distinguish the two types. This
patch splits the old BT_ACL type into two new BT_ACL_IN & BT_ACL_OUT
types.
Change-Id: I7d3c05c26d2a70f80fb1229e245aa21673ec378b
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
The buffer handling code is quite large and will continue to grow. As
it's not strictly HCI core material it's better to just manage it in a
separate file. This patch moves the code to include/bluetooth/buf.h
and net/bluetooth/buf.c.
Change-Id: Ie1ff79ac2cfa132359ce9f7674ff812d34b228aa
Signed-off-by: Johan Hedberg <johan.hedberg@intel.com>
Johan Hedberg