Feature/rmii 100M rt #70

ed-xmos · 2024-11-25T19:10:31Z

Add RMII (2b data @ 50MHz) version of the RT 100M MAC.

https://xmosjira.atlassian.net/browse/ETH-14
https://xmosjira.atlassian.net/browse/ETH-12

…rmii_100m_rt * commit '02343159640e8d46ba929d86c8c37cae0e9c45ac': Update test_support hash Use Xsi.get_xsi_tick_freq_hz() in timeout calculation Update tests/helpers.cmake Use custom test_support branch. Remove hardcoded xsi tick freq from tests Hardcode seed to 1 in test_avb_traffic.py. Update changelog and copyright Run tests with or without --seed based on Jenkinsfile parameter Add comments to shaper Renamed test ids in test_rx_queues.py Modified conftest.py to set the seed Added a --seed option to pytest to allow tests to run with a fixed seed Use custom branch for test_support Get test_shaper passing Fixed _max_mbps calculation in DataLimiter.get_ifg(). Increased END_OF_TEST_TIME to avoid test time outs. Re-enabled RGMII 125MHz group a and b tests Port more tests to xs3 architecture Test for xs3 architecture

…IFG_AS_REF_CLOCK_COUNT_4b, RMII_ETHERNET_IFG_AS_REF_CLOCK_COUNT_1b_NO_TAIL_BYTES and RMII_ETHERNET_IFG_AS_REF_CLOCK_COUNT_1b_TAIL_BYTES defines in the code.

ed-xmos · 2024-12-19T13:20:11Z

doc/rst/lib_ethernet.rst

+   - 7
+   - 3 (1-bit), 2 (4-bit) or 4 (1-bit)
+   - 2
+   - ~TBD k


I have addressed this in my next PR (board support one)

ed-xmos · 2024-12-19T13:29:50Z

examples/test_rmii_rt/CMakeLists.txt

@@ -0,0 +1,29 @@
+cmake_minimum_required(VERSION 3.21)


I think these test apps should be removed. They are for my own dev only and are not documented or helpful to the customer. We have proper test apps and we will have a new documented example when we get new HW.

Yes, makes sense to remove them

ed-xmos · 2024-12-19T13:32:48Z

lib_ethernet/api/ethernet.h

+                          SERVER_INTERFACE(ethernet_tx_if, i_tx_lp[n_tx_lp]), static_const_unsigned_t n_tx_lp,
+                          nullable_streaming_chanend_t c_rx_hp,
+                          nullable_streaming_chanend_t c_tx_hp,
+                          in_port_t p_clk, rmii_data_port_t * unsafe p_rxd, in_port_t p_rxdv,


I'm windoering if I can't remove the unsafe bit. I'll add an issue. Currently it means the call to rmii_ethernet_rt_mac has to be wrapped in unsafe which is not nice.

It'll be great if we can remove unsafe. I noticed that the example in the documentation doesn't say unsafe

Yes, I thought that too. Let me have a look. See #76
I has a very long battle with the XC compiler to get it as far as it :(

ed-xmos · 2024-12-19T13:33:47Z

lib_ethernet/lib_build_info.cmake

@@ -23,5 +23,9 @@ set(LIB_COMPILER_FLAGS_mii_ethernet_rt_mac.xc   ${LIB_COMPILER_FLAGS} -Wno-unusu
 set(LIB_COMPILER_FLAGS_rgmii_buffering.xc       ${LIB_COMPILER_FLAGS} -Wno-unusual-code)
 set(LIB_COMPILER_FLAGS_rgmii_ethernet_mac.xc    ${LIB_COMPILER_FLAGS} -Wno-unusual-code)
 set(LIB_COMPILER_FLAGS_ethernet.xc              ${LIB_COMPILER_FLAGS} -Wno-cast-align)
+set(LIB_COMPILER_FLAGS_rmii_ethernet_rt_mac.xc  ${LIB_COMPILER_FLAGS} -Wno-unusual-code)
+set(LIB_COMPILER_FLAGS_rmii_master.xc           ${LIB_COMPILER_FLAGS} -mdual-issue)


This needs tidying. I'll do that

ed-xmos · 2024-12-19T13:40:01Z

tests/helpers.py

+    """ Returns the time it takes the DUT to process a given frame
+    """
+    if mac == 'standard':
+        return 4000 / phy.get_clock().get_clock_cycle_to_bit_time_ratio()


Not sure what these numbers mean? 4000/6000/2000 - maybe a comment?

Oh, this is not used. I'l remove it. The packet_processing_time() function just before it has these hardcoded numbers. It's supposed to return the processing time of a packet, instead, it ignores the data_bytes argument and returns a fixed number derived from these hardcoded values. I'm not sure why this was written in this way and haven't tried changing it. I'll add a TODO in the code atleast.

ed-xmos · 2024-12-19T13:41:13Z

tests/helpers.py

+
+def get_rmii_4b_port_tx_phy(clk, rxd_4b_port_pin_assignment, verbose=False, test_ctrl=None,
+                          do_timeout=True, complete_fn=None, expect_loopback=True,
+                          dut_exit_time=(50 * px.Xsi.get_xsi_tick_freq_hz())/1e6, initial_delay=(130 * px.Xsi.get_xsi_tick_freq_hz())/1e6):


I think due_exit_time is in microseconds? Be good to comment

Yes, its microseconds. I'll add a comment

ed-xmos · 2024-12-19T13:43:39Z

tests/mii_clock.py

@@ -7,7 +7,7 @@
 class Clock(px.SimThread):

    # Use the values that need to be presented in the RGMII data pins when DV inactive
-    (CLK_125MHz, CLK_25MHz, CLK_2_5MHz) = (0x4, 0x2, 0x0)
+    (CLK_125MHz, CLK_25MHz, CLK_2_5MHz, CLK_50MHz) = (0x4, 0x2, 0x0, 0x1)


Was just wondering if we need to support 5MHz for 10Mb RMII? Not for this PR..

It seems 10mbps is supported using 50MHz.

Hmm, didnt occur to me 10Mbps RMII could be a requirement

ed-xmos · 2024-12-19T13:53:50Z

tests/test_4_1_2.py

@@ -26,8 +27,15 @@ def do_test(capfd, mac, arch, rx_clk, rx_phy, tx_clk, tx_phy, seed):
    if tx_clk.get_rate == Clock.CLK_125MHz:
        max_fragment_len = 142

+    if tx_phy.get_name() == "rmii":
+      min_fragment_length = 16 # https://github.com/xmos/lib_ethernet/issues/73


ed-xmos · 2024-12-19T14:00:27Z

tests/test_rx/test_params.json

+        {"phy":"rmii", "mac":"rt_hp", "arch":["xs3"], "rx_width":"4b_upper", "tx_width":"4b_lower"},
+        {"phy":"rmii", "mac":"rt_hp", "arch":["xs3"], "rx_width":"1b", "tx_width":"4b_lower"},
+
+        {"phy":"rmii", "mac":"rt", "arch":["xs3"], "rx_width":"4b_lower", "tx_width":"4b_upper"},


wow - see what you mean about combinations. I think when we get DP RMII we can focus more on the functionality rather testing different port combos.

ed-xmos · 2024-12-19T14:03:04Z

tests/test_rx_queues.py

@@ -281,6 +291,52 @@ def test_rx_queues(capfd, seed, params):
                    weight_tagged=args.weight_tagged, weight_untagged=args.weight_untagged,
                    max_hp_mbps=100)

+    elif params["phy"] == "rmii":


This (getting the PHY) is repeated quite a lot. Possibly worth being a helper. However, repetition is big thing in this test suite and would be a lot of effort to make more shared.

I agree. Let me see if I can clean it a bit

I've added wrapper functions for rmii tx and rx phy creation. Hopefully that cleans up some repetitive code

ed-xmos · 2024-12-19T14:13:56Z

lib_ethernet/src/rmii_master.xc

+        /* Wait for the start of the packet and timestamp it */
+        unsigned sfd_preamble;
+
+        // Ensure we have sufficiently sized buffer


Debug left in

Thanks. Removed it

…rmii_100m_rt * commit '8475203f4d10bd0748a7e0eb18a4066d9a72f89b': Update lib_ethernet.rst Update lib_ethernet.rst Changelog Fix IFG calculation. Document the reasoning behind the RMII_ETHERNET_IFG_AS_REF_CLOCK_COUNT_4b, RMII_ETHERNET_IFG_AS_REF_CLOCK_COUNT_1b_NO_TAIL_BYTES and RMII_ETHERNET_IFG_AS_REF_CLOCK_COUNT_1b_TAIL_BYTES defines in the code. Fix copyright notice Modify test_4_1_2.py to not test fragments less than 16 bytes in length for RMII

ed-xmos

Looks good. A huge amount of work here - especially the tests.

A few comments in there. I have pushed my updates already.

ed-xmos · 2024-12-19T14:29:03Z

tests/test_rx/src/main.xc


    on tile[0]: filler(0x1111);
    #if ETHERNET_SUPPORT_HP_QUEUES
    on tile[0]: test_rx(i_cfg[0], c_rx_hp, i_loopback, i_ctrl[0]);
    on tile[0]: test_rx_loopback(c_tx_hp, i_loopback);
    #else
+    on tile[0]: filler(0x2222);


My previous comment got lost :(

I am not convinced we are properly testing timing. All other threads need to be run with fast mode (except the thread under test, which is hard as we'd need fast mode options in the main MAC) OR we need to constrain the clock frequency so the max MIPs is capped. Eg. 370 MHz / 5 = 74 MHz. I think we should add a basic test to see if we can support differing tail lengths under constrained MIPS. Doesn't need to be a big test or anything. What do you think? DOesn't need to be for this PR but we could add an issue.

oh no :(. I thought the tests were testing timing properly. The threads under test, which I assume are rmii_master_rx_pins_4/1b, rmii_master_tx_pins, mii_ethernet_filter and mii_ethernet_server) are not in fast mode. Everything declared in the test file, so for test_rx for instance, test_rx(), the 2 filler threads and control() should all be in fast mode, only, I just noticed that the filler() threads randomly go in and out of fast mode (RANDOM_FAST_MODE is set to 1) :(

I'll set RANDOM_FAST_MODE to 0 and test if it meets timing. Would that be alright? or do you think we're still not testing timing properly. I didn't check with core freq set to 370MHz with 5 threads since I was getting weird warning s about 100MHz tick not generated correctly so wasn't sure if it was running ok.

I think enabling fast mode properly is a good step. I am assuming that rx_pins for MII is the toughest case. I think to be absolutely sure we should limit the thread speed. That will result in unpleasant warnings but it's a good way to to set limit the pins threads. It can be a separate test and it would need its own XN. TBH, it could be based on my test code which I took out. I don't mind doing that. You have done quite enough testing for now I think!

…eanup to address review comments

…t into feature/rmii_100m_rt

… fast mode

… right now

ed-xmos added 9 commits November 23, 2024 10:21

Initial API for rmii_rt

0c581b4

WIP RMII RT mac top level

3230356

Working initial port extraction from union

4c0075a

Add compiler warning workaround for RMII

2c58632

Simple test app for RMII ethernet build

a648569

Setup API for rmii_master_init_xx

a99bd72

Initial rmii_master.xc

82d3c1e

Re-use mii_ethernet_filter and mii_ethernet_server in RMII master

bb967cf

rmii_master plumbing and prep

f2dc266

ed-xmos self-assigned this Nov 25, 2024

ed-xmos added 20 commits November 26, 2024 10:55

Initial hardware setup for RMII (untested)

e4feb9c

Source check

d654cfa

Initial chunks for rmii_master_rx_pins

ae53788

remove error handling

4657365

Chunks compiling (use *rxdata)

18571fb

Remove unneeded kernel stack

37683bc

Unify rmii_master_tx_pins

b808ac3

rmii_master_tx_pins impl

41afa75

initial tx_packet from mii

07cdc5d

Building tx (not yet rmii)

415d3b9

Tidy rmii_transmit_packet

baef811

Add 1b rmii skeleton

4f91a4e

Plumb in clock txclk for 1b

fb550af

Label enums for easier compiler comparison

228b101

populate Tx pins for RMII with static inlines

6c3dd7d

Comment about optimisation of clock start/stop

88ead40

WIP MII Rx 4b with comments

277e0ca

Fix crc position in rmii_master_rx_pins_4b

916999c

Do single clock stop (at start) for rmii_transmit_packet_1b

a30f9bf

ed-xmos marked this pull request as ready for review December 19, 2024 09:37

Fix IFG calculation. Document the reasoning behind the RMII_ETHERNET_…

1be46f6

…IFG_AS_REF_CLOCK_COUNT_4b, RMII_ETHERNET_IFG_AS_REF_CLOCK_COUNT_1b_NO_TAIL_BYTES and RMII_ETHERNET_IFG_AS_REF_CLOCK_COUNT_1b_TAIL_BYTES defines in the code.

shuchitak force-pushed the feature/rmii_100m_rt branch from e26e811 to 1be46f6 Compare December 19, 2024 11:20

Changelog

8e29d20

shuchitak self-requested a review December 19, 2024 11:31

ed-xmos commented Dec 19, 2024

View reviewed changes

ed-xmos added 2 commits December 19, 2024 13:26

Update lib_ethernet.rst

8a32aeb

Update lib_ethernet.rst

8475203

ed-xmos commented Dec 19, 2024

View reviewed changes

ed-xmos added 3 commits December 19, 2024 14:15

Tidy lib makefiles

c675d74

Remove Ed's test app

eddd88c

ed-xmos commented Dec 19, 2024

View reviewed changes

Shuchita Khare added 4 commits December 19, 2024 15:36

Added wrapper functions for rmii tx and rx phy creation. Some more cl…

9c1fd7f

…eanup to address review comments

Merge branch 'feature/rmii_100m_rt' of github.com:ed-xmos/lib_etherne…

c330bea

…t into feature/rmii_100m_rt

Set RANDOM_FAST_MODE to 0 so the filler threads are always running in…

70985d1

… fast mode

Added pytest option to specify the test level. Not used by most tests…

c2d6242

… right now

shuchitak force-pushed the feature/rmii_100m_rt branch from 80aa95e to c2d6242 Compare December 20, 2024 08:52

shuchitak approved these changes Dec 20, 2024

View reviewed changes

shuchitak merged commit 471b3b1 into xmos:develop Dec 20, 2024
1 check passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Feature/rmii 100M rt #70

Feature/rmii 100M rt #70

ed-xmos commented Nov 25, 2024 •

edited by shuchitak

Loading

ed-xmos Dec 19, 2024

ed-xmos Dec 19, 2024

shuchitak Dec 19, 2024

ed-xmos Dec 19, 2024

shuchitak Dec 19, 2024

ed-xmos Dec 19, 2024 •

edited

Loading

ed-xmos Dec 19, 2024

ed-xmos Dec 19, 2024

shuchitak Dec 19, 2024

ed-xmos Dec 19, 2024

shuchitak Dec 19, 2024

ed-xmos Dec 19, 2024

ed-xmos Dec 19, 2024

shuchitak Dec 19, 2024

ed-xmos Dec 19, 2024

ed-xmos Dec 19, 2024

shuchitak Dec 19, 2024

ed-xmos Dec 19, 2024

shuchitak Dec 19, 2024

shuchitak Dec 19, 2024 •

edited

Loading

ed-xmos Dec 19, 2024

shuchitak Dec 19, 2024

ed-xmos left a comment

ed-xmos Dec 19, 2024 •

edited

Loading

shuchitak Dec 19, 2024

ed-xmos Dec 19, 2024 •

edited

Loading

Feature/rmii 100M rt #70

Feature/rmii 100M rt #70

Conversation

ed-xmos commented Nov 25, 2024 • edited by shuchitak Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ed-xmos Dec 19, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

shuchitak Dec 19, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ed-xmos left a comment

Choose a reason for hiding this comment

ed-xmos Dec 19, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ed-xmos Dec 19, 2024 • edited Loading

Choose a reason for hiding this comment

ed-xmos commented Nov 25, 2024 •

edited by shuchitak

Loading

ed-xmos Dec 19, 2024 •

edited

Loading

shuchitak Dec 19, 2024 •

edited

Loading

ed-xmos Dec 19, 2024 •

edited

Loading

ed-xmos Dec 19, 2024 •

edited

Loading