04:14 <ehmry> I wrote an Nix package for mirage, I will try to wrap it to build unikernels later today https://github.com/ehmry/mirage-nix

09:22 <samoht[m]> rgrinberg: I've created https://gist.github.com/samoht/6c96e2a73c45c90e3fda3a23e249a582 to show an example of a dune workflow with the mirage tool

09:38 <samoht[m]> https://gist.github.com/samoht/6c96e2a73c45c90e3fda3a23e249a582#file-dune-L7-L27 can be auto-generated by interpretating config.ml (so it needs to compile it first). In the currnet patches, `mirage configure` generates the first stanza and fork-exec to `./config.exe configure` which is not terrible.

09:51 <samoht[m]> There is also now a `mirage query` command which allows to query the dynamic parameters (just the list of libraries atm, but I'm adding the list of files for the configure and build steps today)

09:52 <samoht[m]> (it also fork-exec to `./config.exe query <...>` so it needs to compile config.ml)

10:47 <hannes_qubes> so pv with minios-master and a hello-world unikernel (in debug mode): https://paste.x25519.net/DWT#G8bqsNZ6NJdVdj7zYu0NMNmkoNZd+OOE2SrwxrAKlSI=

10:47 <hannes_qubes> pvh same unikernel https://paste.x25519.net/JZi#y/1xLFy2juNtLZpspbVWbSOeYotP/9jnHZtdVCyN3aE=

10:47 <hannes_qubes> marmarek: ^^

10:48 <mato[m]> hannes_qubes: Test

10:48 <hannes_qubes> mato[m]: yes, i can see your test.

10:49 <mato[m]> hannes_qubes: Working on the jackline packaging again (have now figured out how to make it work on Ubuntu)

10:49 <mato[m]> hannes_qubes: However, due to mirage-crypto releases master does not build for me anymore. Do you have a "cheat sheet" of what I need to change in jackline to get it to work?

10:49 <hannes_qubes> ...and somehow the entire get_console/xenstore_start_page C code in mirage-xen is never executed, but it fails before

10:50 <hannes_qubes> mato[m]: no cheat sheet, I intended to work on this later today (porting applications / figuring out what to do). reynir intended to work on some cheat sheet yesterday (not sure whether he did)

10:51 <mato[m]> hannes_qubes: in the interim, is there something i can do in the jackline opam file to force it to depend on the old crypto?

10:52 <hannes_qubes> mato[m]: i guess otr < 0.3.7 should do the trick (maybe x509 < 0.10.0 and tls < 0.11.0 is needed as well)

10:52 <mato[m]> thanks, i'll try that

10:53 <hannes_qubes> base is 0x65676172696d looks a bit suspicious (that's even ascii ;), and page fault at 0 looks like a null pointer.. but hannes doesn't know much C.

10:59 <mato[m]> hannes: actually, adding a "conflicts:" for "mirage-crypto" does the trick (at least in terms of "opam install --deps-only ." doing the right thing)

11:26 <marmarek> hannes: yes, this address is definitely suspicious

11:27 <marmarek> hannes: do you have quick copy&paste instruction how to build this binary from your branches?

11:28 <marmarek> the good think you have anything on the console, otherwise it would be much harder :)

11:35 <hannes> marmarek: https://paste.x25519.net/j5G#AiRHncTgdNQwCMCUNQGrT4gUSdILleFb18J3llSIZRQ=

11:35 <hannes> I have not tested that, if you run into compilation issues, please complain

11:38 <hannes> marmarek: and thanks to printk I figured the code in mirage-xen start_info_stubs is never executed.. I manually checked the initialisation sequence (from mini-os/kernel.c and mirage-xen/bindings/main.c -- this looks fine to me)

11:52 <marmarek> hannes: [WARNING] Failed checks on minios-xen package definition from source at git+https://github.com/mirage/mini-os.git#minios-master-with-pvh:

11:52 <marmarek> error 57: Synopsis and description must not be both empty

11:52 <marmarek> is it an issue?

11:54 <talex5_> No, that's just a warning.

11:56 <talex5_> I have Hannes's branches building and running (up to the page fault) on my Qubes machine now too.

11:57 <tg> o/

11:57 <tg> greets to all in marrakech and not marrakech

11:58 <marmarek> $ opam install mirage-xen

11:58 <marmarek> [mirage-xen.3.2.0] no changes from git+https://github.com/mirage/mirage-xen.git#pvh

11:58 <marmarek> <><> Synchronising pinned packages ><><><><><><><><><><><><><><><><><><><><><><>

11:58 <marmarek> No solution found, exiting

11:58 <marmarek> Sorry, no solution found: there seems to be a problem with your request.

11:59 <marmarek> --debug output don't mean much to me

12:00 <tg> what's the status of the wireguard implementation?

12:00 <tg> seen this repo by janestreet

12:01 <tg> hannes ?

12:01 <tg> https://github.com/janestreet/noise-wireguard-ocaml/ -- this one i mean

12:04 <tg> and also, anyone knows about this kv store if it's usable yet and if works with mirage? https://github.com/tomjridge/tjr_kv http://www.tom-ridge.com/2019-08-21_ml_workshop_kv_store_for_ocaml.html

12:06 <talex5_> marmarek: this worked for me: https://paste.x25519.net/DKb?hl=dockerfile#UaACcMuuvrfdOWJY43HJBri5FRCq8MRbsl5wQ1HAKxI=

12:11 <talex5_> I'm a bit surprised it picked 3.2.0 as the default version. Maybe you need to `opam update` first?

12:11 <marmarek> maybe ocaml version? `opam switch list-available` doesn't list 4.08 for me

12:14 <talex5_> Yeah, sounds like you need to "opam update".

12:23 <marmarek> better

12:33 <marmarek> got the page fault

12:39 <tg> and is anyone working on a QUIC impl in ocaml?

12:44 <tg> mato: any idea about unikernels in rust?

12:46 <tg> mato: have you looked into running that on solo5?

12:47 <hannes> marmarek / talex5: my current suspicion is that nobody calls the get_xenbus () -- when the mirage-bootvar-xen still contained the xenstore lookup, it sometimes just halted without a page fault... but yeah, happy if any of you take it from here :) it is indeed very convenient to have printk around... it may as well be that some mirageos-minios patches are missing (but I went throug that list multiple times

12:47 <hannes> (and didn't pick the arm related patches))

12:48 <hannes> tg: i do not know. you could investigate by trying it out and reporting back whether it works.

12:49 <marmarek> hannes: the PF seems to be on caml_start_program() _returning_

12:49 <marmarek> so the actual error is somewhere earlier

12:49 <tg> ok i'll check, thought you might know more since you committed something in there.. otherwise the repo looks not really active

12:50 <tg> mato: so far found libhermit-rs with uhyve but dont really know details.. https://github.com/hermitcore/libhermit-rs

12:56 <hannes> marmarek: hmmmmmm

12:58 <marmarek> caml_program is NULL

12:58 <marmarek> in caml_start_program

12:58 <marmarek> some linking failure?

13:00 <marmarek> $ objdump -t hello.xen|grep caml_program

13:00 <marmarek> 0000000000000000 g F .text0000000000000670 caml_program

13:02 <marmarek> do you have PV binary handy? can you check that objdump on it?

13:08 <hannes> i don't, i have lunch (and bootet my other operating system)

13:08 <hannes> but uh-oh, thanks for looking into this :)

13:26 <marmarek> built PV one

13:26 <marmarek> $ objdump -t hello.xen|grep caml_program

13:26 <marmarek> 0000000000003000 g F .text0000000000000670 caml_program

13:46 <marmarek> where caml_program come from?

13:52 <talex5_> I think that might be the actual OCaml program that you compile, rather than a C function.

13:59 <talex5_> `objdump -d hello.xen` shows that `caml_program` is indeed at address 0.

14:12 <marmarek> hmm, gdb says nothing is mapped at 0

14:12 <marmarek> maybe some restriction about mapping initial pages?

14:14 <marmarek> is there some linker script I can modify to place it somewhere else?

14:15 <talex5_> mini-os's `x86/minios-x86_64.lds` maybe?

14:15 <hannes> marmarek: i think the linker script is in mini-os arch/x86/minios-x86.lds.S

14:18 <marmarek> interesting, output should start with .bootcode and only then .text - so if caml_program is at 0x0, it means .bootcode is empty?

14:20 <marmarek> ".section .bootcode" is under #ifdef CONFIG_PARAVIRT

14:34 <h01ger> morroco has now imposed a departure and arrival ban for plans coming and going abroad. a bit sooner than i expected/feared 4 days ago :/

14:34 <h01ger> (source https://www.spiegel.de/wissenschaft/medizin/coronavirus-news-am-sonntag-die-wichtigsten-entwicklungen-zu-covid-19-und-sars-cov-2-a-637532e2-ed16-4e84-9c46-c17886861684 )

14:36 <hannes> h01ger: thanks, but tbh yes, we're reading the news as well. no, i don't see any reason to discuss this (at length) in this channel. yes, we're ok and we'll be ok.

14:38 <hannes> marmarek: i remember we had a bootcode patch in our mini-os patch series, but upstream already had some .bootcode in their linker script (those under the ifdef CONFIG_PARAVIRT) -> i thought pvh may not need them

14:39 <marmarek> hannes: interestingly, at the very beginning 0x0 is properly mapped and I see caml_program there

14:39 <marmarek> but at the time caml_start_program() is called, it is unmapped

14:40 <talex5_> Is it just zeroes there?

14:40 <h01ger> hannes: fully agreed. i'm glad this channel is mostly technical and fun to watch!

14:40 <marmarek> no, nothing - looks like present=0 in respective page table entry

14:40 <marmarek> (trying to access page tables through gdb now)

14:41 <talex5_> I guess I'd expect something to map out page zero to detect null-pointer errors.

14:42 <marmarek> that would be my guess too

14:47 <talex5_> Hmm, mini-os doesn't compile with MM_DEBUG defined.

14:52 <hannes> can we just start with some other address, not 0x0 somehow in the linker script, then?

14:59 <talex5_> Hmm: /* Use first page as the CoW zero page */

15:01 <marmarek> I've moved it at 0x1000

15:02 <marmarek> and now got #GP

15:10 <marmarek> ok, instead of starting everything at 0x1000, I've added 0-filled page at 0x0

15:10 <marmarek> and not it is much better

15:11 <marmarek> https://gist.github.com/marmarek/39d7ff8f1b67a0ee3b3528b088a292d2

15:12 <talex5_> That looks much better to me :-)

15:12 <talex5_> What is your linker script now?

15:12 <marmarek> added patch there

15:13 <marmarek> that isn't nice method of doing it, because it adds all those zero-es literally to resulting binary

15:15 <marmarek> I've reused the fact that .bootcode is placed at the beginning

15:23 <talex5_> mini-os's gettimeofday was called and returned 1584285707 s + 662831 us, which looks sensible.

15:23 <talex5_> But: Fatal error: exception Failure("unix_gettimeofday returned number of days outside int range")

15:30 <hannes> i think we can handle 4k 0s ;)

15:30 <hannes> good work, thanks

15:49 <mato[m]> If you figure out what the problem(s) are, I can help massage the ldscript into a "better" version, BTDT

15:50 <mato[m]> See e.g. Solo5 ldscripts which I have extensively commented (so as not to forget what I was doing and why)

16:23 <mato[m]> hannes: By the way, you may have missed my earlier comment -- adding a "conflicts: [ mirage-crypto ]" is sufficient to get jackline to build in the presence of the new crypto releases. This may be useful for other packages in the interim.

16:24 <hannes> mato[m]: cool, thanks. i also ported jackline to mirage-crypto, found an issue in otr, and re-released otr.

17:10 <hannes> ^ is merged to master now, so if you rebase you should not need any conflict anymore :)

17:24 <mato[m]> hannes: will do in a bit. just cleaning up the work in progress commits I did earlier today.

17:45 <fhpedro> what's the meaning of "Mirage" name

17:45 <fhpedro> ?

17:58 <reynir> hannes, mato, here's what I've been writing so far regarding nocrypto -> mirage-crypto migration: https://gist.github.com/reynir/f68eee2de94fbebf0701bef7a99b9140