kern/108581: [sysctl] sysctl: hw.acpi.cpu.cx_lowest: Invalid argument

[Bruce Cran]

On Thu, 26 Mar 2009 09:37:50 -0400
John Baldwin <jhb at freebsd.org> wrote:

> On Wednesday 25 March 2009 6:39:14 pm Bruce Cran wrote:
> > On Fri, 20 Mar 2009 00:30:03 GMT
> > Daniel Dvořák <dandee at hellteam.net> wrote:
> > 
> > > The following reply was made to PR kern/108581; it has been noted
> > > by GNATS.
> > > 
> > > From: =?UTF-8?Q?Daniel_Dvo=C5=99=C3=A1k?= <dandee at hellteam.net>
> > > To: <bug-followup at FreeBSD.org>,
> > > 	<lars.stokholm at gmail.com>
> > > Cc:  
> > > Subject: Re: kern/108581: [sysctl] sysctl: hw.acpi.cpu.cx_lowest:
> > > Invalid argument Date: Fri, 20 Mar 2009 01:01:51 +0100
> > > 
> > >  This is a multi-part message in MIME format.
> > >  
> > >  ------=_NextPart_000_0007_01C9A8F7.746C4190
> > >  Content-Type: text/plain;
> > >  	charset="UTF-8"
> > >  Content-Transfer-Encoding: quoted-printable
> > >  
> > >  Hi acpi team,
> > >  =20
> > >  today I have installed fbsd 7.1R on one box with this relativly
> > > old = error and I was surprised about results .. it is the same:
> > >  =20
> > >  # uname -a
> > >  FreeBSD X.Y.Z 7.1-RELEASE FreeBSD 7.1-RELEASE #0: Thu Jan  1
> > > 14:37:25 = UTC 2009
> > > root at logan.cse.buffalo.edu:/usr/obj/usr/src/sys/GENERIC  = i386
> > >  
> > >  # sysctl dev.cpu.0.cx_supported
> > >  dev.cpu.0.cx_supported: C1/0
> > >  
> > >  # sysctl hw.acpi.cpu.cx_lowest=3DC1
> > >  hw.acpi.cpu.cx_lowest: C1
> > >  sysctl: hw.acpi.cpu.cx_lowest: Invalid argument
> > >  =20
> > >  # sysctl hw.acpi.cpu.cx_lowest=3DC0
> > >  hw.acpi.cpu.cx_lowest: C1
> > >  sysctl: hw.acpi.cpu.cx_lowest: Invalid argument
> > >  =20
> > >  # sysctl hw.acpi.cpu.cx_lowest=3DC1/0
> > >  hw.acpi.cpu.cx_lowest: C1
> > >  sysctl: hw.acpi.cpu.cx_lowest: Invalid argument
> > >  
> > >  # dmesg -a | grep "acpi"
> > >  acpi0: <ASUS P4S8X-X> on motherboard
> > >  acpi0: Overriding SCI Interrupt from IRQ 9 to IRQ 20
> > >  acpi0: [ITHREAD]
> > >  acpi0: Power Button (fixed)
> > >  acpi0: reservation of 0, a0000 (3) failed
> > >  acpi0: reservation of 100000, ff00000 (3) failed
> > >  acpi_timer0: <24-bit timer at 3.579545MHz> port 0xe408-0xe40b on
> > > acpi0 acpi_button0: <Power Button> on acpi0
> > >  pcib0: <ACPI Host-PCI bridge> port 0xcf8-0xcff on acpi0
> > >  atkbdc0: <Keyboard controller (i8042)> port 0x60,0x64 irq 1 on
> > > acpi0 cpu0: <ACPI CPU> on acpi0
> > >  hw.acpi.cpu.cx_lowest:
> > >  hw.acpi.cpu.cx_lowest
> > 
> > I think I've found the problem and have updated the PR kern/108581
> > (http://www.freebsd.org/cgi/query-pr.cgi?pr=kern/108581). The global
> > cpu_cx_count was being initialized to 0 in acpi_cpu_startup
> > (in /sys/dev/acpica/acpi_cpu.c) but code below it appears to assume
> > that it's been intialized to 3 because it only sets it if it's
> > higher than the current CPU supports - that is, cpu_cx_count should
> > reflect the highest Cx state that all CPUs support.
> > 
> > There's also a bug in the _CST section just below it; I think the
> > line:
> > 
> > if (sc->cpu_cx_count > cpu_cx_count)
> > 
> > should be
> > 
> > if (sc->cpu_cx_count < cpu_cx_count)
> 
> No, the code is doing things differently on purpose (though I'm not
> completely sure why).  For _CST it sets cpu_cx_count to the maximum
> Cx level supported by any CPU in the system.  For non-_CST it sets it
> to the maximum Cx level supported by all CPUs in the system.  I think
> it is correct for cpu_cx_count to always start at 0 and only be
> bumped up to a higher setting.  Setting it to 3 would be very wrong
> for the _CST case as I've seen CPUs that support C4.
> 
> Note that C1 _always_ exists as it is simply the "hlt" instruction
> that has existed since the 8086.  Only C2+ require power-saving
> extension support in the CPU, so cpu_cx_count should always end up >=
> 1.  It would be interesting if you could add some debug printfs to
> print out the values that acpi_cpu_generic_cx_probe() computes for
> 'sc->cpu_cx_count' (sysctl dev.cpu could be useful for this) as well
> as all changes to the 'cpu_cx_count' global variable.
> 

For my Athlon XP CPU, acpi_cpu_generic_cx_probe sets sc->cpu_cx_count
to 1, and subsequently dev.cpu.0.cx_lowest has always worked.  After
adding printfs I found that the problem is that the cpu_cx_generic
block in acpi_cpu_startup is being run and because cpu_cx_count is set
to 0 it never gets updated; the statement "if (sc->cpu_cx_count <
cpu_cx_count)" is never true.

-- 
Bruce Cran