00:16
sandeepkr has quit [Read error: Connection reset by peer]
00:33
sandeepkr has joined #m-labs
00:45
<
GitHub172 >
buildbot-config/master caf951b Sebastien Bourdeauducq: remove migen -> misoc build trigger
01:04
bb-m-labs has quit [Quit: buildmaster reconfigured: bot disconnecting]
01:05
<
GitHub80 >
buildbot-config/master 7967634 Sebastien Bourdeauducq: only run gateware simulations on Linux and exclude them from coverage
01:05
<
GitHub80 >
buildbot-config/master 1b65b5f Sebastien Bourdeauducq: do not install Migen and MiSoC for Windows tests
01:05
bb-m-labs has joined #m-labs
01:05
<
GitHub >
artiq/master 657afd7 Sebastien Bourdeauducq: artiq/test/gateware -> artiq/gateware/test...
01:25
<
whitequark >
sb0: what are next priorities?
01:31
<
sb0 >
ffs had a flterm running
01:32
<
GitHub >
artiq/master 43aad09 Sebastien Bourdeauducq: python3.5 -> python3...
01:41
<
sb0 >
I suppose #638 is also quick, and then we can release 2.2
01:42
<
whitequark >
is that the only thing blocking it?
02:15
bb-m-labs has quit [Quit: buildmaster reconfigured: bot disconnecting]
02:15
<
GitHub130 >
buildbot-config/master f2e1f06 Sebastien Bourdeauducq: fix 7967634
02:16
bb-m-labs has joined #m-labs
02:16
<
GitHub144 >
buildbot-config/master 03d30ef Sebastien Bourdeauducq: fix syntax error
02:16
<
sb0 >
bb-m-labs, force build artiq
02:16
<
bb-m-labs >
The build has been queued, I'll give a shout when it starts
02:16
<
bb-m-labs >
build #1263 forced
02:16
<
bb-m-labs >
I'll give a shout when the build finishes
02:46
<
sb0 >
fails with test_rpc_timing on win64
02:46
<
sb0 >
whitequark, have you tried having more network buffers?
02:46
<
sb0 >
*packet buffers
03:07
<
GitHub >
artiq/master f6024b6 Sebastien Bourdeauducq: drtio: fix ad9154 extension registration
03:07
<
GitHub >
artiq/master 3aced46 Sebastien Bourdeauducq: dashboard: report lost connection with master. Closes #602
03:07
<
GitHub >
artiq/master dcea48a Sebastien Bourdeauducq: protocols/broadcast,sync_struct: add disconnect callback
03:33
kuldeep has joined #m-labs
03:33
sandeepkr has quit [Read error: Connection reset by peer]
03:33
kuldeep_ has quit [Read error: Connection reset by peer]
04:33
<
whitequark >
sb0: not yet
04:33
<
whitequark >
I should finish that too
04:34
<
sb0 >
bb-m-labs, force build migen
04:34
<
bb-m-labs >
build #130 forced
04:34
<
bb-m-labs >
I'll give a shout when the build finishes
04:40
bb-m-labs has quit [Quit: buildmaster reconfigured: bot disconnecting]
04:40
<
GitHub4 >
buildbot-config/master 7589d0e Sebastien Bourdeauducq: install Python 3.5 for migen
04:40
bb-m-labs has joined #m-labs
04:41
<
sb0 >
bb-m-labs, force build migen
04:41
<
bb-m-labs >
build #131 forced
04:41
<
bb-m-labs >
I'll give a shout when the build finishes
04:50
<
GitHub162 >
buildbot-config/master 9212e70 Sebastien Bourdeauducq: iverilog is not used anywhere
05:02
<
GitHub >
artiq/master 7daab07 Sebastien Bourdeauducq: drtio: fix syntax/import
05:22
sandeepkr has joined #m-labs
05:40
<
GitHub >
artiq/master bff5f1b Sebastien Bourdeauducq: conda: update migen
05:46
<
sb0 >
taking the drtio transceivers to 3gbps is, of course, another shit show
05:55
<
sb0 >
_florent_, any update on that awful packet corruption bug?
06:03
<
sb0 >
oh, of course, I forgot to update the si5324 settings
06:03
<
sb0 >
whitequark, do you still have their tool installed?
06:04
<
whitequark >
sb0: yeah
06:04
<
whitequark >
which settings do you want me to run through it?
06:05
<
sb0 >
150MHz and 50MHz
07:21
sandeepkr has quit [Ping timeout: 252 seconds]
07:22
<
GitHub31 >
misoc/master b0d940a whitequark: liteeth: make rx/tx slot count configurable.
07:25
<
whitequark >
sb0: in si5324.rs you're using a whole lot of return Err
07:25
<
whitequark >
for this kind of thing panics are preferable
07:26
<
whitequark >
to be more specific: in map_frequency_settings.
07:26
<
whitequark >
the rest is fine
07:30
<
GitHub28 >
misoc/master 02884fe whitequark: kc705: forward rx/tx slot count to LiteEthMAC from MiniSoC.
07:42
<
GitHub >
artiq/master b95db4f whitequark: Use four ethmac buffers instead of two....
07:42
<
GitHub >
artiq/master 3a19a9f whitequark: firmware: minor cleanup in board::si5324.
08:04
<
GitHub >
artiq/master c585784 whitequark: conda: bump misoc dependency.
08:07
kuldeep has quit [Remote host closed the connection]
08:08
kuldeep has joined #m-labs
08:55
<
sb0 >
whitequark, can i use the kc705s?
08:55
<
whitequark >
sb0: yes
08:55
<
whitequark >
btw, enlarging the buffers worked pretty well
08:56
<
whitequark >
there are still retransmits, of course, but the situation is significantly less bad
08:56
<
whitequark >
ideally what should happen is rolling back buffers to 2x and then getting data into/out of them by an interrupt
08:56
<
sb0 >
how many retransmits are there compared to lwip?
08:57
<
whitequark >
not sure.
08:57
<
whitequark >
does it matter? I was going to implement an atomic ringbuffer anyway, for the UART
08:58
<
sb0 >
an atomic ringbuffer?
08:58
<
whitequark >
a ringbuffer with atomic push/pop operations
08:58
<
whitequark >
for use from interrupts.
08:58
<
whitequark >
well, not just atomic but lock-free.
08:59
<
sb0 >
well, it matters because there are other things to do, like DMA and this getattr bug
08:59
<
sb0 >
a lwip level of retransmits is acceptable for now
09:13
<
sb0 >
whitequark, is that 200ms just for the TCP communications?
09:17
<
whitequark >
sb0: no
09:17
<
whitequark >
200ms is the retransmit time.
09:17
<
whitequark >
the rest is insignificant
09:18
<
whitequark >
to be specific, it takes 200ms to retransmit and the rest of communication, handling, etc is 27ms
09:20
<
whitequark >
lol what the actual hell
09:20
<
whitequark >
the
*entire first packet burst* is just zeroes
09:21
<
whitequark >
ok oh not entire. ELF padding before .text
09:23
<
sb0 >
but after the first retransmit, the PC calms down and the next kernels on the same connection don't incur another 200ms?
09:24
<
whitequark >
I don't think so, no
09:25
<
sb0 >
okay, good to know
09:25
<
whitequark >
I was unclear. I don't think it calms down.
09:26
<
whitequark >
anyway this ring buffer should not be complex to write at all, I'll hopefully do it next morning
09:26
<
whitequark >
just four CASes
09:26
<
sb0 >
how large is the TCP window?
09:26
<
sb0 >
how many packets will the PC send without waiting for the first ack?
09:26
<
whitequark >
64k, same size as the underlying buffer
09:27
<
whitequark >
the problem with setting the TCP window based on ethmac buffer state is it doesn't play well if we have multiple sockets open
09:28
<
whitequark >
and it also doesn't play well with chatter on the interface, e.g. broadcast ARP, ICMPv6, etc
09:28
<
sb0 >
so the ISR would copy the packets into some socket-allocated buffer?
09:28
<
whitequark >
a separate buffer
09:28
<
sb0 >
or some buffer large enough for a lot of connections/chatter?
09:28
<
sb0 >
(some global buffer)
09:29
<
whitequark >
it has to be separate because smoltcp isn't mt-safe/reentrant
09:29
<
whitequark >
you can only really poll it in the main loop, and we don't have any upper bound on the time between polls
09:30
<
whitequark >
actually, I wonder how this affects #407
09:39
<
sb0 >
even with the proper si5324 settings 3gbps is still the same shit
09:42
sandeepkr has joined #m-labs
09:46
kuldeep has quit [Remote host closed the connection]
09:47
kuldeep has joined #m-labs
09:59
<
whitequark >
sb0: so about #638
09:59
<
whitequark >
the root cause is that methods defined
*only* on instances are not supported, for a good reason
10:01
<
whitequark >
mh, I guess I could add a special case...
10:10
<
GitHub >
artiq/master 2f37b1d whitequark: compiler: support methods defined on singleton instances....
10:22
<
whitequark >
sb0: hm, what's a function like hasattr() but without considering inherited/class fields?
10:25
<
GitHub >
artiq/master 6acdcbb whitequark: Fix 2f37b1d.
10:42
<
GitHub187 >
misoc/master 421b197 whitequark: Fix the value of mem::ETHMAC_SIZE.
10:44
<
whitequark >
sb0: what are arb_req and arb_gnt dma csrs?
10:50
<
sb0 >
whitequark, assert arb_req, then wait until you get arb_gnt. then you are permanently controlling (D)RTIO until you deassert arb_req.
10:56
<
whitequark >
hmm, what's up with the windows rpc rtt...
17:51
mumptai has joined #m-labs
19:41
_rht has joined #m-labs
22:49
<
sb0 >
whitequark, why didn't you simply use a boost converter for the ion gauge hv power supply?
22:52
<
sb0 >
couple hundred volts is doable without having to wind a transformer
23:39
mumptai has quit [Remote host closed the connection]
23:42
<
sb0 >
rjo, do we really need a 50T FPGA on Kasli?
23:42
<
sb0 >
and we don't need a si5324 in standalone mode, do we?
23:48
sandeepkr has quit [Ping timeout: 255 seconds]
23:51
<
rjo >
sb0: well a spartan6 45 is a very tight fit for two cpus. i figured a 50t would be barely ok.
23:52
<
rjo >
sb0: i'd like to not change kasli too much but alter the design so that it can be run stand-alone. i don't want to add another board to the ecosystem. that's inefficient.
23:56
<
sb0 >
we can also DNP stuff or use pin-compatible FPGAs, but I'm not sure if it makes sense at those volumes
23:57
<
sb0 >
is it the two CPUs that take all the space in the pipistrello? I thought it was the FIFOs
23:58
<
sb0 >
plus SERDES logic