Rev |
Log message |
Author |
Age |
Path |
24 |
Added build scripts for KC705 (somehow they got omitted in version 22) |
wzab |
3596d 17h |
/fade_ether_protocol/ |
23 |
Added script allowing for batch compilation of FADE 10G for Atlys.
Just run build_proj_atlys.sh
Removed compilation of receiver2.c (which has been deleted) in the Makefile.
Slightly changed order of operation in the receiver2t.c |
wzab |
3596d 17h |
/fade_ether_protocol/ |
22 |
Sources reorganized so, that each ".xci" is in a separate directory.
Added script for rebuilding of Vivado project for KC705.
Just run the build_proj_kc705 script, and the project
will be created as kc705_10g3.xpr in the kc705_10g3 drirectory.
You can open it later on in Vivado 2014.3 and build the bitstream. |
wzab |
3596d 18h |
/fade_ether_protocol/ |
21 |
Changed the top entity for KC705, so that the FADE core is correctly reset
by the RESETMAC command.
Changed sending of the FCMD_RESET command form driver to the FPGA (as it
is not confirmed - the reinitialized core can't confirm the command).
Changed the receiver2t demo program so that it sends RESETMAC three times
before the transmission is started.
In the future, probably there should be a STATUS ioctl, which would
verify, that the core has been reset by RESETMAC.
Then the program should send RESETMAC, and afterwards read the STATUS,
verifying, that RESETMAC arrived FPGA. If not it should repeat the
above procedure (not implemented yet). |
wzab |
3597d 01h |
/fade_ether_protocol/ |
20 |
Cured problem with dta_eod missed, when buffer was full at the moment
when dta_eod was asserted.
It caused the client program to hang, as no end of data was transmitted
and no new data arrived... |
wzab |
3597d 16h |
/fade_ether_protocol/ |
19 |
The send_cmd function protected to attempt to call it if the Ethernet device was not
claimed before (previously it caused a problem with NULL pointer dereference
in the ioctl in kernel code).
Moved the Ethernet device name to a macro, so it is easier to modify it, when
someone tests in different machines (e.g. the 10Gb/s NIC for tests with
KC705 is p4p1, while 1Gb/s NIC for tests with Atlys is eth0) |
wzab |
3597d 18h |
/fade_ether_protocol/ |
18 |
Changed handling of commands so, that all commands are now confirmed - even start and stop
(which previously were not confirmed - successful reception of start was confirmed
by sending of data packets (what if there was no data at the moment?), successful
reception of stop was confirmed by end of data stream).
Added also flushing of data stream after reception of stop command.
The receiver2t.c implements suggested method of communication with the FADE driver.
One thread is responsible for reception and processing of data, while another thread
is responsible for sending of control commands. |
wzab |
3598d 14h |
/fade_ether_protocol/ |
17 |
Corrected mutex management in ioctl. |
wzab |
3604d 18h |
/fade_ether_protocol/ |
16 |
Introduced inline functions simplifying reading and writing of packets' contents. |
wzab |
3604d 19h |
/fade_ether_protocol/ |
15 |
The previous experimental implemenation for 10Gb/s links has been replaced with the
new, rewritten implementation, which uses sequential numbering of packets instaed
of "sets of packets". This allowed to simplify the driver, and improve its performance.
Additionally a possibility to send user defined commands to the FPGA was added. |
wzab |
3612d 17h |
/fade_ether_protocol/ |
14 |
Removed small mistake causing a combinatorial loop. |
wzab |
3617d 18h |
/fade_ether_protocol/ |
13 |
Small correction in the kernel driver |
wzab |
3620d 00h |
/fade_ether_protocol/ |
12 |
Added experimental version of my protocol, working with the 10Gb/s
link on the KC705 board.
The design has been initially tested, and is working, but it still
needs some improvments.
Please note, that this version uses the "jumbo" frames,
with 8192 bytes of user data in each packet (plus header).
Therefore it was necessary to make some changes in the kernel
driver. |
wzab |
3626d 20h |
/fade_ether_protocol/ |
11 |
Corrected small bug in eth_sender8 (which in fact should only
generate a compiler warning, as the resulting bit vector
was anyway correctly truncated...) |
wzab |
3783d 18h |
/fade_ether_protocol/ |
10 |
Updated test client program - receiver2.c
Added possibility to activate only selected slave in the command line.
Currently program services up to three slaves: 0, 1 and 2
To activate a slave, add its number as a program argument in the command line
(order doesn't matter). |
wzab |
4083d 16h |
/fade_ether_protocol/ |
9 |
Corrected serious bug in acknowledged packed number calculations, which lead to transmission
errors in high speed CPUs. |
wzab |
4084d 16h |
/fade_ether_protocol/ |
8 |
Adjusted parameters of the NCA (network congestion avoidance) alogorithm
to obtain higher throughput. |
wzab |
4088d 17h |
/fade_ether_protocol/ |
7 |
Shortened synchronization path for tx_start signal in the eth_sender blocks |
wzab |
4088d 20h |
/fade_ether_protocol/ |
6 |
Further adjusted IFG lengths |
wzab |
4089d 03h |
/fade_ether_protocol/ |
5 |
Decreased length of the intraframe gaps (IFG)
In 100Mb/s version - to 100bits (25 nibbles)
In 1Gb/s version - to 104bits (13 bytes) |
wzab |
4089d 18h |
/fade_ether_protocol/ |