Monday, 2022-03-28

Veera[m]lkcl: Hi! I can't git push to libreriscv.git repo though pull is working07:57
programmerjakeVeera: see
Veera[m]programmerjake: read the msg can you renable my write access08:22
programmerjakesorry, i can't, luke is the only one with the access to change write access08:23
programmerjakewe'll have to wait for him08:23
lkclVeera[m], programmerjake is taking responsibility for giving you the instructions needed. please ask him what you need to do.10:05
programmerjakeVeera: run `git branch --contains 3fda698000d39457eed8f5a17a2c6980958ceacb`10:10
programmerjakeif the output contains `master`, then we'll have to rebase your commits10:11
programmerjakeif it doesn't, then you should be fine10:12
Veera[m]ran it does not contains 'master'10:13
programmerjakewhat about `git branch --contains 1d17f24414c4b4e7ebf2d06d30c21f9938d60d09`?10:14
Veera[m]ran. it says 'no such commit'10:15
programmerjakeok, luke, please re-enable Veera's git access, his repo's fine10:15
programmerjakelkcl: ^10:16
lkclok great. 1 se11:08
lkclVeera[m], programmerjake, done11:08
programmerjakethx. sorry for the trouble.11:11
lkcli have 3 days in which to get FPGA demo running11:13
lkcli haven't even started on ethernet or spi flash11:13
lkcli've only just managed to get HyperRAM operational and only the lower 8 mbytes11:13
programmerjakehmm...the ecp5 has enough block ram that you could probably have some kind of rtos in it that you could use as a hyperram, ddr3, or spi flash needed...idk if that meets the requirements though11:17
lkclnot in the slightest.11:18
lkclthe project is to deliver a router product. that requires OpenWRT.11:18
lkclthat in turn requires running a linux kernel11:19
lkclwith 2 months spent not being able to get DDR3 memory beyond Icarus Simulation i had to turn, in an emergency, to HyperRAM11:19
lkclha, hilarious. i had all 32 MBytes accessible yesterday evening15:08
lkclbut because i had written the test program to index by *words*, i thought "err there's 4x less *bytes* here"15:09
zemaye_has the idea of continuous integration builds been brought up? I think a lot of the devscript stuff this weekend would have been caught by it18:28
markostrying to follow the docs,
markossv.ld 10.v, 0(in).v seems to be valid according to this18:54
markos(where in is: .set in, 5)18:54
markosbut I'm getting a assembler Error: unrecognized opcode: `sv.ld'`18:57
markossetvl seems to be recognized just fine and replaced by a .long equivalent18:57
markosalso, the tool seems to also replace even commented out #sv.* instructions18:59
markoslkcl, ^18:59
tplatenI'm trying on the verilator backend, external_core_top.v is ignored for verilator builds, it is only used for FPGAs19:26
tplatenI've pushed my changes to tplaten_3d_game, I get an error in 13. Executing Verilog backend.19:37
tplatenERROR: Non-constant enable $memwr$\memory$external_core_top.v:282587$25668_EN in memory initialization $meminit$\memory$external_core_top.v:282587$25707. (with fpga-target=verilator)19:38
tplatenWhen building for the orangecrab, it takes much more time until I get the unconstrained in LPF error (6 minutes), so I first try to get libre-soc+microwatt peripherials working in verilator.20:13
tplatenI had a look at the microwatt_verilator branch, it only supports EXTERNAL_CORE for fpga targets such as the ulx3s, for verilator you can only use the microwatt core, I was expecting you can use either core.20:16
lkclmarkos, yes anything prefixed "#sv.", the program's not hugely sophisticated.  put "#  # sv." i.e. some spaces in and it'll "go away".21:09
lkclline 6621:09
lkclthere you go.21:10
lkclfirst identify anything starting "#"21:10
lkclah spaces are not sufficient (due to the strip)21:11
lkclput "# commented out sv."21:11
lkclline 118621:14
lkcl1185         'sv.stw 5.v, 4(1.v)',21:15
lkcl1186         'sv.ld 5.v, 4(1.v)',21:15
lkclthose are examples of working syntax21:15
lkcland there are plenty more 1 sec21:15
lkclif i can find the damn things )21:16
lkclokaay here we go21:18
programmerjakezemaye_: yes, we had been working on getting CI working, but it stalled... i have a computer for CI sitting in my room basically all set up but luke wanted CI build logs to be available on a libre-soc mailing list or git or something, rather than only on debian salsa (the CI server is a gitlab runner configured to use debian salsa as the managing gitlab instance), we never finished getting it set up due to22:14
programmerjakedisagreements/misunderstandings on who needed to do what....22:14
programmerjakewe've been working on other stuff and never got around to finishing getting it working22:15
lkclyeah there's nothing stopping you from going ahead and setting up CI, jacob23:07
lkclwhat i'm not going to do however is set up a mailing list on the libre-soc server that spews out tens to hundreds of thousands of lines of log files per run, overwhelming the Virtual Machine's 10 GB storage within months23:08
lkcl(and i'm not prepared to pay more for it either)23:08
lkclthat does not stop *you* from setting up mailman - or other mailing list - on the CI server23:08
lkcland i believe in the bugreport discussion it says - or should say - that i'm happy to do "redirection".23:09
lkclthe CI machine is, i believe, already joined to the VPN running on libre-soc so it already has access to SMTP via
lkclstorage of hundreds of [irrelevant] gigabytes of logs on your *local* machine is then possible23:10
lkclif any of those logs become important then it is a trivial matter of copying the relevant sections to a bugreport.23:11
zemaye_This is interesting. I have to pop offline.23:15
programmerjakewell...with git doing compression into packfiles (all logs go into git, the mailing list only has 1-2kB per build), i expect it to be on the order of a few tens/hundreds of kB per build.23:18
programmerjakei don't have hundreds of GB of storage in the CI server...iirc it has a 64GB ssd and that's it.23:18
programmerjakeso that's around 2GB of git repo per year assuming 50 CI builds per day (50 is kinda excessive...our average rate is much lower)23:23
programmerjakeand around 40MB of mailing list23:23
programmerjake-- totally affordable imho23:23
lkcla 64 GB SSD is still 6x larger than the spare capacity of the VM. so, no, sorry.23:25
lkclplus i have to do remote scp backups onto my laptop (rsync)23:26
lkclstorage of repeated logs "everything's ok" is not useful - at all.23:26
programmerjakemost of the 64GB is needed for OS and temporary storage,23:26
lkclwe can always get you a bigger SSD/HDD23:27
lkclthe VM has to be paid for - by me - on a monthly basis.23:27
lkclextra storage has to be paid for at a cost per megabyte.23:27
programmerjakealso, the CI server will have availability issues due to being on my local network23:28
lkclperfectly fine23:28
lkclit's a "non-essential" resource23:28
lkclif it can't send out mail to anyone who's subscribed to its list, it will wait until it's back online and re-send23:29
lkclonce it gets its OpenVPN link restored23:29
programmerjakealso, a mailing list is a poor method of handling build emails...ideally we'd send emails to those who broke the build, rather than spamming everyone everytime anyone commits anything23:32
lkclif that's possible, that's great. it can be done via the (open) SMTP port on
lkclif the connection was reliable i'd recommend ssmtp23:34
lkclhowever if it's not going to be reliable then store-and-forward will be needed, and the default exim4 debian install should be fine, setting as the smtp forwarding host23:35
programmerjakei'd use postfix since i'm waay more familiar with it23:35
lkclconfigure it to "pretend" to be and by the time the mail gets to the server it'll *be* @libre-soc.org23:35
lkclthere's no smtp user/password needed for anything on 10.6.0.*23:36
programmerjakeok, so i guess i'll work on that today...23:38
programmerjakei don't have a public ip, so i'll need some https forwarding set up so people can access the git repo and mailman interface, we also need some kind of allowlist of who we can send build failure emails to, otherwise we'll spam random people if we do something like push gcc's source to one of our repos23:43
lkclthat should be perfectly fine, if you can work out the nginx snippet needed for a virtual host (or, simpler, a directory-transparent-proxy) i'll add it.23:46
lkcla virtual host means i have to add a DNS record as well, which is extra work23:47
programmerjakeimho a virtual host is probably best, avoiding the build server accidentally messing up the main website...may i suggest
programmerjakeor build1/libre-soc.org23:52
lkclalain's busy running a training course at the moment, he should be done by the end of the week23:56

Generated by 2.17.1 by Marius Gedminas - find it at!