There is non negligible time between the moment init_timeslot is
called and the moment when we get the first RX (which is when
we start sending out data at the same rate we RX it). And those
1024 byte were borderline not enough to not underflow.
By delaying the 1024 byte pre-fill until we get the first RX,
we ensure the time that pre-fill covers is more consistant and
less dependend on whatever the app/scheduling does during the
init phase.
Signed-off-by: Sylvain Munaut <tnt@246tNt.com>
Change-Id: Ic1c93fd138073a75830dc16bb41c4541e68eef90
Until the remote side sends the first byte, we send fill
data since it can take it some time after the mode change
and the file descriptor is open for data to show up.
Previously due to a bug, the warning about short read was
not printed, but now that it's fixed, you get a bunch of
fairly "useless" warning at the beginning of the timeslot.
This fixes that since reported data by _e1_tx_raw is "faked"
until the pipe is active. (Fill data is actually added upstream)
Signed-off-by: Sylvain Munaut <tnt@246tNt.com>
Change-Id: Iaa6268e9b4a7bce5ea3027a29c48c147499373be
No functional changes, just making things a bit more organized
for the next patch adding an _e1_tx_raw.
The order after the patch is :
_e1_tx_hdlcfs
_e1_ts_read
_e1_line_mux_out_channelized
_e1_line_mux_out_superchan
e1_line_mux_out
_e1_rx_raw
_e1_rx_hdlcfs
_e1_ts_write
_e1_line_demux_in_channelized
_e1_line_demux_in_superchan
_e1_line_demux_in_ts0
e1_line_demux_in
Signed-off-by: Sylvain Munaut <tnt@246tNt.com>
Change-Id: I2ced3a5ba3cf3cc1f212d8f3e46cfefa15e4af0c
A lot of them are related to signedness or type range limitation.
A lot are not actual issues and work find in practice, but a few
lead to actual bad behavior.
This makes all the conversion explicit to mark intent.
Signed-off-by: Sylvain Munaut <tnt@246tNt.com>
Change-Id: I992b9bc51659e85966651b1146091501b8f149f5
Go with -Wextra but removing some of the obnoxious options it brings with
it.
Signed-off-by: Sylvain Munaut <tnt@246tNt.com>
Change-Id: Iaafd35ba39e272d5900b31b14d2651e0ecb8a84a
Newer descriptors have the max packet size set to 3 to save on
isochronous bandwidth.
By specs, the feed back endpoints transfer are 24 bits anyway,
so no point is using anything larger.
Signed-off-by: Sylvain Munaut <tnt@246tNt.com>
Change-Id: I09e259db539e62571fd17fc16ffaa317564ba257
In osmo-e1-hardware.git Change-Id Ic4f57cf79bd32cf75f81ef3073cb8d4a2d1857d8
we added support for passing RAI (remote alarm indication) as a flag via
the USB interrupt messages; let's add support for this here.
osmo-e1d already internally parses TS0 to determine the same
information, and we have to keep this for backwards-compatibility with
older firmware builds. But maybe at some future point we can remove
our own TS0 scanning code here and rely on the USB device to inform
us about remote alarms.
Change-Id: Ie1994968e792c37f9272b9854547db95a41cab5b
Some time ago we added various error counters, but for two of them
we missed to actually ever increment them at all.
Change-Id: Ieb83a2e2e83e334c543bee83726f04f83b19227a
We just found a bug in icE1usb (likely firmware) which was hard to find
as there was zero notification from osmo-e1d that it actually never
received any data from the icE1usb hardware/firmware anymore.
Change-Id: Id22e4110b9067f50b1818eb12295b2d4eb9cdc12
This avoids truncation and the following error message when operating
in HDLC-FCS mode:
<0001> mux_demux.c:91 (I0:L0:T16) Truncated message: Client tried to send 320 bytes but our buffer is limited to 264
Change-Id: I836d8b98ce5b831b0498c4650263ec3b3d4f2c45
This allows us to reliably detect client disconnection at least in
the case of RAW mode channels. Even with this patch applied, e1d
still fails to reliably detect client disconnect on HDLC-FCS channels.
Change-Id: Ifb8b91d39b394f9c10c859f3adac85ea47b7653f
There was some wrong ordering in config_write_e1d() regarding
resolving the vpair 'peer' devices. We can only call
e1d_vpair_intf_peer() _after_ we have established that the
given interface actually is of type VPAIR.
Assert failed intf->drv == E1_DRIVER_VPAIR vpair.c:96
Change-Id: If494d77ed1df5cda655d3b4a60868154dc2b355e
osmo-e1gen is a program that re-uses large parts of osmo-e1d, but whose
main purpose is to generate a variety of error conditions in order to
test a remote E1 implementation.
Instead of using the automatisms of the icE1usb transmit IP core, it
switches it to transparent mode and uses a host-software based E1 framer
"osmo_e1f", over which we have more control than the firmware.
Change-Id: I53a86d6730eb76a9cff9eb3f4786139015c91230
The remaining intf_line.c really only manages the data structures.
This is useful for building other programs than osmo-e1d, such as
an upcoming E1 test utility called osmo-e1gen which will also use
the USB interface and icE1usb hardware, but not any of the mux/demux/ctl
code.
Change-Id: I1ceaea85a15e2fae1d2e041173be9d758c6d0b78
Experimentally this has shown a much lower probability of
USB related overruns/underruns, at least on my system.
Change-Id: I5c73b29462a4870d3c86fd7f46e1574daae6f93f
prior to this patch, the configured vpair interfaces would not
be saved to the config file on 'write file'.
Change-Id: Iff6551318534a3717c374060082147f17b925a21
Align the behavior of osmo-e1d with that of DAHDI: If a timeslot is
opened once, it cannot be opened again by anyone until it is closed
by the current owner. This way we'd have the same failure semantics
in DAHDI vs. osmo-e1d, which is very useful in case of misconfiguration
when osmo-bsc + osmo-mgw would "fight" over a timeslot.
Add a osmo_e1dp_client_ts_open_force() function that allows to override
and get back the original behavior.
Closes: OS#4654
Change-Id: Ib25adf827ec41e74de15e0e4fdcfc9bcc9a32e58
every so often, the USB transfer completes without data (due to
"unlucky" time alignment between E1 and USB frame clock). Don't
call the demuxer in that case.
Otherwise the user is confused by error messages like
<0001> intf_line.c:467 (I0:L0) IN ERROR: -4
Change-Id: Ia99f97c2cca44d15a83a54cebe884b343ec44f46
When opening a timeslot, the client can now specify the (from
application perspective) timeslot read buffer size. This breaks
ABI, but then we're still at a point where only prototype hardware
exists, so we can get away with it.
Change-Id: I6d603778cce14c5d72fe5f54904905ea7e66d7ff
Existing applications (such as those written for DAHDI) expect to
be reading data in buffer/chunk sizes. For example OsmoNITB: It
doesn't want to execute an expensive read/recv syscall to receive 11
bytes, if it needs at least 160 bytes.
Change-Id: I807671bc6f2acaef740ce215b8d8abcb5dce2640
This groups all HDLC-specific members together, in preparation
of adding more fields for other modes.
Change-Id: Ide0577c25836252862153b4f24da550bee013687
This adds the related code to the server and client side of the CTL
interface to switch a line between CHANNELIZED and SUPERCHANNEL.
Change-Id: I765b5c3bc9e07b2353f8647e8260ff95df3727e6
Sylvain Munaut
Harald Welte
Pau Espin
Oliver Smith