2020-11-04 Meeting Meeting notes

Owned by Mike Holmes
Last updated: Nov 06, 2020
3 min read

Meeting Details

Topic: Linaro Open Discussions - Kernel related
Time: Nov 4, 2020 02:00 PM London

Join Zoom Meeting
https://linaro-org.zoom.us/j/98027304997

Meeting ID: 980 2730 4997
One tap mobile
+16699009128,,98027304997# US (San Jose)
+12532158782,,98027304997# US (Tacoma)

Dial by your location
+1 669 900 9128 US (San Jose)
+1 253 215 8782 US (Tacoma)
+1 301 715 8592 US (Germantown)
+1 312 626 6799 US (Chicago)
+1 346 248 7799 US (Houston)
+1 646 558 8656 US (New York)
888 788 0099 US Toll-free
877 853 5247 US Toll-free
Meeting ID: 980 2730 4997
Find your local number: https://linaro-org.zoom.us/u/azkgHLreP


Attendees

  • Mike Holmes

  • Jammy

  • Linus
  • Lorenzo (arm)
  • Sami (arm)
  • Shameerali (Huawei)
  • Sudeep (arm)
  • Vincent
  • Hanjun
  • Jean-Phillip
  • Viresh
  • Zhangfei (Linaro)
  • Jonathoan (Huawei)
  • Don
  • Sheirlock (Huawei)
  • Mattero
  • Arnd

  • WangZhou

  • Ulf

Agenda

  • Hanjun/Shameer - IORT reserved memory support
  • Zhangfei/Wangzhou - SVA support for SMMU stall mode
  • Hanjun - Uncore DVFS and how to support it (needs spec update, either ARM specs or ACPI)
  • Mike Holmes Gague interest in repeating this call


Under discussion for inclusion

  • Matteo Carlini - Generic Initiator Proximity Domains in ACPI - Patches merged
  • Hanjun - MPAM status and anything we can do to speed up the progress Dropped after the discussion on the list
  • Jonathan - IOMMU performance work (may need Will Deacon involved from Google)
  • Jonathan - vCPU hotplug follow up after the KVM forum session

  • IORT
    • Shameer:   node support - a way to describe memory regions, use case card has FW that needs to allocate host memory. using SMMU no longer needed with IORT
    • - patches out for kernel, and ACPI, got a mail from Steve (Arm) 
    • is IORT version final, ACPI folks asking.
    • LP: to check if it is final : JC: Arm usually good at versioning. Sami asked for one minor change to IORT spec.
    • SKT: AMD can specify flags, unity mapped, the possibility of holes in VM address space
    • LP: are you asking to add something, SKT no not at this time LP: it is worth asking
    • LP: what happened with kexec, what happens to the memory, things should be brought back, Steve is looking at this.
    • LP: What happens if the kernel resets, JC: ACPI has this, you would need to scan the existing settings.
    • LP: could be very complicated, in ACPI can force kernel not to reassign: JC is it "should not"
    • JP: need to walk it to solve both cases, but nasty.
    • LP: X86 has preserved things, it may just work but likely to have a problem, JP: not assured to happen
    • two items to look at
      • bdf matching ? → not just a Linux thing, 
      • kexec → on the list
    • JC : don't want to see this patch stall whilst working the other issues
    • LP: bdf should not be a blocker, but we should address it
    • JP : for 1.0 print warning if RM
  • SVA
    • discussed with Arm before (JP), violation, need suggestions
    • LP: 
    • JP: was designed correctly at the time of the spec
    • LP: for follow calls we need to support the call with a couple of slides have seen patches on the list. Don't think patches will fly if the next generation doesn't need it what then
    • JP: I don't think we have the right people, need PCI Sig  JC: don't think we need to this Sig, it is a tiny code fix. JP: I think we were planning this fix.
    • JC: Is it ok to have quirk fix, but maintainer asked to find a proper fix.
    • LP: need SMMU support needed first
    • JC: PCI maintainer needed to talk long term fix
    • LP: What else, not complicated, just controversial
    • JP: virtual SVA, Eric Auger looking at this (not on call)
    • LP: there is DT story, be good to have it JC: we don't have a DT so that is hard
    • JC: no worries if Arm propose patches with DT
    • JP: 
  • DVFS
    • can use MMIO, PSCI or ACPI (mailbox)
    • Do we have any direction for the solution
    • VG: for the uncore is it about power or frequency, power down or idle.  Have you considered SCMI interface?  Hanjun Only supports device tree, 
    • SH: do you have to have the CPU track Uncore freq : HJ: yes : SH: can you have firmware do it ? JC: not that simple because PCI devices can affect it.
    • VG: is the bandwidth affected? " JC: yes: VG: we have this form mobile, you say you have BW requirement. JC: can we use SCMI ?  SH: We have to make it work from the spec
    • SH: does PCI also need fine-grain control, we might need to add to the spec.
    • SH: What are the requirements? HJ: Now we just frequency scale, 
    • JC: in this case PCI is not enough, it is fiddly to determine.
    • HC: We need to gather all the info, need to talk in SWG about DVFS
    • JC: would like ACPI to be able to describe the system but unlikely to happen.
    • JC: standard SW will not happen any time soon

Action items

  •