IRC channel logs

<black_knight>Good *, regarding the discussion you had yesterday, if I understand correctly, you still need to fix apic support to enable smp and x64.. is this far down the road? How could one contribute?

<damo22>we need libacpica to compile as a userspace library

<damo22>then we can link it to the acpi translator to fetch the PCI pin mappings

<damo22>i have done some work on this

<black_knight>Where is the source code for this?

<black_knight>I found this exchange https://www.mail-archive.com/bug-hurd@gnu.org/msg31613.html is this up to date?

<damo22> http://git.zammit.org/hurd-sv.git/?h=acpica-integrate

<damo22>we probably dont want to import acpica code into hurd, rather push the needed tweaks upstream to acpica

<damo22>and then use it as libacpica.a

<youpi>yes

<damo22>but i needed to try it first so i imported it to a branch i was working on

<black_knight>Why is this necessary for smp? Isn’t this about power states?

<black_knight>maybe you need that to properly shutdown the rump disks? I guess

<youpi>acpi is a beast

<youpi>everything goes through it nowadays

<youpi>irq numbering, smp cores enumeration, etc.

<black_knight>Ah so you are building an IO server that work as an accountant for who is using what resource and acpi provides the tables needed for that.

<black_knight>i thought you already had that

<damo22>i just built a minimal layer to request irq pin mappings

<damo22>but it also does shutdown

<damo22>we need a way to call ACPI functions

<damo22>so i made it call into libacpica

<damo22>for example, you cant initiate Sx power states without an ACPI parser these days

<damo22>the hack i have currently only works on QEMU

<damo22>and we cant turn on SMP with rumpdisk because it doesnt know which IRQ to use yet

<damo22>or im not sure if SMP is tied to APICs

<damo22>i think you need to turn off the legacy PIC to have SMP, and that means setting up IOAPIC IRQs correctly

<damo22>hence the need for ACPI parser to get the mappings

<black_knight>I think I’m missing what is the basic architecture: I guess the IDT is set by gnumach, how is the Irq delivered from the gnumach gate to rumpdisk?

<youpi>the interrupt handler queues an RPC message

<black_knight>What is the interface to set those messages?

<youpi>see device_intr_notify in include/device/notify.defs

<youpi>and device_intr_register / ack in device.Defs

<black_knight>I’m looking this up, thanks.. Zheng Ada work

<youpi>yes. Note that we reworked it a bit

<akib>ping Pellescours

<Pellescours>pong

<akib>does hurd support smp?

<Pellescours>as said before, there is some parts but it was untested and it’s not finished

<Pellescours>damo22: I tried to build hurd against upstream rump and I have 2 problems

<akib>is it needed to specify cpu count at compile time?

<akib>what problems???

<Pellescours>it’s is able to detect the number of cpus but I don’t know if it’s able to run multiple cpus

<Pellescours>1. generate the rump headers, I saw the function generatekernelheaders in the buildrump.sh but I have difficulties to generate them

<akib>where's the source?

<Pellescours>it’s on git https://savannah.gnu.org/projects/hurd

<akib>which branch?

<Pellescours>master

<damo22>oh the headers, i didnt try to get that far

<damo22>i just linked the binary libs

<damo22>the smp part is in gnumach mostly i think

<damo22>oh hes gone

<damo22>i gtg to sleep

<Pellescours>good night

<damo22>thanks

<akib>pmy: hi

<ArneBab>I just tried to install git and mercurial in a Hurd, but I get the error that they both require more up to date versions of dependencies:

<ArneBab> git : Depends: git-man (< 1:2.32.0-.) but 1:2.33.0-1 is to be installed

<ArneBab> mercurial : Depends: mercurial-common (= 5.6.1-4) but it is not going to be installed

<ArneBab>Is there an easy fix?

IRC channel logs

2021-09-11.log