cvs commit: src/sys/boot/pc32/libi386 biosacpi.c src/sys/conf acpi.mk files src/sys/dev/acpica5 Makefile Makefile.inc acdragonfly.h acpi.c acpi_acad.c acpi_button.c acpi_cmbat.c acpi_cpu.c acpi_ec.c acpi_lid.c acpi_resource.c acpi_thermal.c ...

[YONETANI Tomokazu]

On Thu, Jan 18, 2007 at 02:56:36PM +0000, Johannes Hofmann wrote:
> YONETANI Tomokazu <qhwt+dfly at les.ath.cx> wrote:
> > On Thu, Jan 18, 2007 at 06:40:05AM +0000, Johannes Hofmann wrote:
> >> YONETANI Tomokazu <qhwt+dfly at les.ath.cx> wrote:
> >> > On Wed, Jan 17, 2007 at 11:18:06PM +0100, Michal Belczyk wrote:
> >> >> On Wed, Jan 17, 2007 at 08:42:49PM +0000, Johannes Hofmann wrote:
> >> >> > Sorry for not testing earlier, but with this commit my thinkpad T42p
> >> >> > no longer boots with acpi enabled. It spits out some warnings and then
> >> >> > hangs (see http://www.ecademix.com/JohannesHofmann/tmp/acpi.jpg for
> >> >> > a screenshot).
> >> > 
> >> > What was the next few lines after "acpi0: <IBM TP-1R> on ..." before?
> >> > No ctrl+alt+esc?
> >> > 
> >> >> Same here on T60.
> >> > 
> >> > Please let me know the result of either or both of the following
> >> > intructions:
> > [snip]
> >> >  it should print the function names on the screen, I'd like to know
> >> >  the last few ones.
> >> 
> >> http://www.ecademix.com/JohannesHofmann/tmp/acpi2.jpg
> > 
> > That's somewhere around here in acpi_attach; AcpiInitializeSubsystem()
> > has ACPI_FUNCTION_TRACE but it's not shown in acpi2.jpg, so it's stuck
> > after entering acpi_attach() but before calling AcpiInitializeSubsystem().
> > Can you insert a few kprintf()'s in the section below to see where it's
> > actually stuck?
> > 
> > Cheers.
> > 
> > 
> > [acpi_attach:acpi.c]
> > 
> >      ACPI_FUNCTION_TRACE((char *)(uintptr_t)__func__);
> >  #if __FreeBSD_version >= 500000
> >      /* Initialise the ACPI mutex */
> >      mtx_init(&acpi_mutex, "ACPI global lock", NULL, MTX_DEF);
> >  #endif
> > <<<<<< FROM HERE >>>>>>
> >      ACPI_LOCK;
> >      sc = device_get_softc(dev);
> >      bzero(sc, sizeof(*sc));
> >      sc->acpi_dev = dev;
> >      callout_init(&sc->acpi_sleep_timer);
> > 
> >      /*
> >       * Set the globals from our tunables.  This is needed because ACPI-CA
> >       * uses UINT8 for some values and we have no tunable_byte.
> >       */
> >      AcpiGbl_AllMethodsSerialized = (UINT8)acpi_serialize_methods;
> >      AcpiGbl_EnableInterpreterSlack = TRUE;
> > 
> >      error = ENXIO;
> >  #ifdef ACPI_DEBUGGER
> >      debugpoint = kgetenv("debug.acpi.debugger");
> >      if (debugpoint) {
> >          if (!strcmp(debugpoint, "tables"))
> >              acpi_EnterDebugger();
> >          freeenv(debugpoint);
> >      }
> >  #endif
> > <<<<<< UNTIL HERE >>>>>>
> >      if (ACPI_FAILURE(status = AcpiInitializeSubsystem())) {
> >          kprintf("ACPI: initialisation failed: %s\n",
> >                 AcpiFormatException(status));
> >          goto out;
> >      }
> > 
> 
> 
> By adding kprintfs() I traced it down to hang somewhere in
> nsload.c:AcpiNsLoadTable.
> But I think there should have been some trace lines if it comes that far.

Hmm, db_print_backtrace() should print the backtrace, but it may not be
very useful if you can't use the keyboard and the trace is too deep.

> So I suspect that there is something wrong with the debugging code.
> Does function tracing work for you?

Yes, with or without ACPI_DEBUG_{LOCKS,MEMMAP}
(but the message buffer is too short to hold the whole trace).
  http://les.ath.cx/DragonFly/dmesg.boot

Cheers.