Part_blk/USB issue

[Norman Feske]

Hi Ben,

please don't feel offended when a particular post of your's does not
receive immediate attention. Most of the regular developers want to give
profound advice rather than shooting in the blue. For coming up with a
good advise, however, one must take some time to investigate. This is
particularly the case for assisting with practical troubleshooting,
which often requires experimentation. Given the fact that each of us
works quite intensively on topics different from your's, it is not
always possible to prioritize your questions over the current line of
work. Sometimes, a meaningful response may take some days. Should a
question remain unanswered for a week, a wake-up message from your side
is appreciated. ;-)

On 01/06/2015 02:55 AM, Nobody III wrote:
> I have been experimenting with things related to my issue when using
> part_blk with ahci, and I have found that it works fine in both QEMU and
> Virtualbox. However, when I combine part_blk with usb_drv, I encounter
> what appears to be the exact same problem, but it only occurs on some of
> the kernels. The scenario works bug-free on okl4 and fiasco, but freezes
> on foc and nova. Note that other servers such as rump_fs work just fine
> with the usb driver, so the problem isn't a simple driver failure. I've
> attached the run script for the usb and part_blk scenario. I hope this
> helps fix my AHCI issue (which only occurs on real hardware) as well.

Good that you could draw the line between these kernels. In fact, OKL4
and L4/Fiasco solely rely on the legacy PIC (programmable interrupt
controller) whereas Fiasco.OC and NOVA use the APIC. When using the APIC
however, the interrupt numbers found in the PCI devices do not
correspond to the numbers used (by the kernel and core's IRQ service) to
program the interrupt controller. The numbers must be translated
according to the information found in the ACPI tables. For this reason,
drivers must use the IRQ service provided by the ACPI server instead of
the IRQ service provided by core.

In your run script, the default route prefers parent services over child
services. Hence, when the USB driver opens an IRQ session, it gets
routed to core's IRQ service. The driver looks up the IRQ number of the
USB host controller by looking into the controller's PCI configuration
space and then opens an IRQ session at core with the untranslated
number. Eventually, the driver listens to the wrong interrupt. Try
equipping the USB driver with the following route:

  <route>
    <service name="IRQ"> <child name="acpi" /> </service>
    <any-service> <parent /> <any-child /> </any-service>
  </route>

This way, IRQ session requests from the USB driver will go to the ACPI
server, which applies the necessary translation of interrupt numbers.
Under the hood, the ACPI server opens an IRQ session at core, but using
the translated number. For more information, there is a README file for
the ACPI driver:


https://github.com/genodelabs/genode/blob/master/repos/os/src/drivers/acpi/README

Cheers
Norman

-- 
Dr.-Ing. Norman Feske
Genode Labs

http://www.genode-labs.com · http://genode.org

Genode Labs GmbH · Amtsgericht Dresden · HRB 28424 · Sitz Dresden
Geschäftsführer: Dr.-Ing. Norman Feske, Christian Helmuth