The original trxcon_fsm I wrote back in 2017 [1] was more like
a boolean flag, as there were only two states: IDLE and MANAGED.
Not surprising, given that until recently handling of multiple
L1CTL connections was not supported. Now that we have this
implemented, lifetime of a trxcon_fsm instance is limited by
lifetime of a L1CTL connection, what renders the FSM useless.
This change removes the old 'boolean' trxcon_fsm and introduces
the new one, which will allows us to abstract the L1CTL interface
from the TRXC/TRXD interfaces, as well as the scheduler. The new
FSM will also simplify development of the RLC/MAC layer for GPRS.
Change-Id: Ifaf63ead9dd180181358e771367b2a686ba159ca
Related: [1] I7ee6fc891abe5f775f5b7ebbf093181a97950dea
Related: OS#5599
This change makes it possible to configure l1sched related logging:
* l1sched_logging_init() allows to configure logging categories for
common and data messages (by default, DLGLOBAL is used);
* struct l1sched_cfg allows to configure a logging prefix to be
used in l1sched messges (by default, 'l1sched[0x%p] is used)'.
Let's use osmo_fsm_inst_name(trxcon->fi) as the logging prefix.
Change-Id: I26da1a506b02502a3a6a887533c35fb09c13c429
Related: OS#5599, OS#3761
l1ctl_server_start() does not actually start the server, it simply
allocates memory and initializes it. l1ctl_server_shutdown() does
the opposite. Let's use more precise symbol names.
Change-Id: Ie039abdff3911c5b566c760b26c31203824c5764
Calling l1ctl_server_shutdown() whenever the server is not initialized
will result in accessing uninitialized values. This can happen if we
goto exit before l1ctl_server_start() was called.
Initializing the server with zeroes is not an option, because we need
to initilize llist_head and osmo_fd structures with proper values.
Allocate and initialize struct l1ctl_server in l1ctl_server_start(),
deinitialize and free() in l1ctl_server_shutdown().
Change-Id: Idf13914fd0b0ae09b2ce5ece1f4203ebcae05d6e
Related: CID#275254
Ideally, FBSB procedure should be implemented as a state of trxcon's
FSM. For now let's simply move the related fields to trxcon_inst.
Remove l1ctl_shutdown_cb() as it's not needed anymore.
Change-Id: I92b50cf1bb36886fbe3d6460af3c0b1c57256ae8
trxcon consists of the following three main components:
* the L1 TDMA scheduler (libl1sched),
* L1 interface (TRXC/TRXD over UDP),
* L2 interface (L1CTL).
In [1] and [2] the L1 scheduler was abstracted out from both L1
and L2 interfaces and separated into a library, so it does not
use the TRXC/TRXD nor L1CTL related API directly.
This change is the next step towards the goal of having all three
components abstracted from each other. Moreover, this patch brings
us closer to another goal of being able to support multiple L1CTL
connections (each having its own scheduler).
The idea is to give both L1 and L2 interfaces access to the
'trxcon_inst' structure, which basically groups all three components
mentioned above into a single piece. The end goal is to eliminate
direct interaction between the interfaces, and the scheduler, so that
one could easily replace TRXC/TRXD and/or L1CTL with something else.
Change-Id: I23319951c56577085e1092669b5534f9d6bda48d
Related: [1] I31f77976a7a225ef292fe6dcd583513aec97ed44
Related: [2] I001fb7bc2663eea308b5a8882746ed9863f2c2f8
Using L1CTL specific structures as the primitive payload was a
beautiful hack in the early days of trxcon. But since we're
going to separate the scheduler into an interface independent
library, we have to introduce and use an abstract API.
Change-Id: I84597d44ea7d74b8840a919ecb09988ba1980a73
Related: OS#5599, OS#3761
Make l1sched_prim_alloc() private and call it from l1sched_prim_push().
This makes the API more convinient, because both functions are always
used together.
Change-Id: Ia9c0170fb06efcef569e987b57ab9ab7f7c7e847
Related: OS#5599, OS#3761
The current function name is confusing, because l1sched_prim_init()
is not only initializing a primitive, but also allocating it on heap.
Let's use '_alloc' instead of '_init' to reflect that.
Change-Id: Ie1bebb6829ba9f640455685fcd7309b6aa442ef0
Related: OS#5599, OS#3761
This is the first step towards the goal of moving the scheduler
into a separate library.
Change-Id: Ifa6137c239c215a3d323213ee74d34b419622be4
Related: OS#5599, OS#3761
Vadim Yanitskiy