DragonFly bugs List (threaded) for 2007-02
[
Date Prev][
Date Next]
[
Thread Prev][
Thread Next]
[
Date Index][
Thread Index]
Re: [issue557] 1.9.0-DEVELOPMENT hangs with ACPI enabled
On Thu, Feb 22, 2007 at 08:46:42AM -0300, Mariano Aliaga wrote:
> >Hmm.. what happens if you undo the patch, and change above code to:
> >%%%%
> > acpi_id = obj->Processor.ProcId;
> >kprintf("Before calling AcpiOsFree()\n");
> > AcpiOsFree(obj);
> >kprintf("After calling AcpiOsFree()\n");
> > if (acpi_pcpu_get_id(device_get_unit(dev), &acpi_id, &cpu_id) !=
> > 0) {
> >kprintf("acpi_pcpu_get_id() failed\n");
> > return (ENXIO);
> > }
> >kprintf("acpi_id: %d, cpu_id: %d\n", acpi_id, cpu_id);
> >
> >%%%%
> >without debug.acpi.layer or debug.acpi.level in /boot/loader.conf.
> >
>
> This is what I get: http://www.e-nix.com.ar/~maliaga/00002.jpg
Probably I should've asked earlier; if you booted FreeBSD, or DragonFly
built from source code before import of new ACPI-CA, do you have
hw.acpi.cpu sysctl knob? The picture is somewhat overwrapped, but it
seems that acpi_pcpu_get_id() failed for both units. You saw 4
"acpi_pcpu_get_id() failed" message, right? I think the lock-up occurred
somewhere other than acpi_cpu_probe(), after returning from it, but
whether it locks up or not depends on what acpi_cpu_probe() does.
I wrote another patch to find it. Please apply attached patch after
reverting the previous change, and install the driver. Boot into the
boot loader prompt, and type(without other debug.acpi variables set)
the following two lines to boot into single-user mode:
set debug.acpi.cpu.knob=5
boot -sv
if it locks up, write down(or take a picture of) lines containing "ProcId",
which looks like this:
cpu0: ProcId 1: acpi_id=0, cpu_id=0
or
cpu0: ProcId 1: can't find cpu_id
Then reboot into the boot loader, and this time try 4 instead of 5.
Repeat this until it stops locking up (or when the number reached to 0),
and tell me the number.
(you don't need to write down the "ProcId" line; only the first time).
Cheers.
Index: acpi_cpu.c
===================================================================
RCS file: /home/source/dragonfly/cvs/src/sys/dev/acpica5/acpi_cpu.c,v
retrieving revision 1.19
diff -u -p -r1.19 acpi_cpu.c
--- acpi_cpu.c 22 Feb 2007 04:02:50 -0000 1.19
+++ acpi_cpu.c 24 Feb 2007 11:07:43 -0000
@@ -182,10 +182,15 @@ static devclass_t acpi_cpu_devclass;
DRIVER_MODULE(cpu, acpi, acpi_cpu_driver, acpi_cpu_devclass, 0, 0);
MODULE_DEPEND(cpu, acpi, 1, 1, 1);
+static int acpi_cpu_probe_debug_knob = 0;
+TUNABLE_INT("debug.acpi.cpu.knob", &acpi_cpu_probe_debug_knob);
+#define DEBUG_KNOB(n) \
+ if (acpi_cpu_probe_debug_knob == (n)) return(ENXIO)
+
static int
acpi_cpu_probe(device_t dev)
{
- int acpi_id, cpu_id, cx_count;
+ int acpi_id, cpu_id, proc_id, cx_count;
ACPI_BUFFER buf;
ACPI_HANDLE handle;
char msg[32];
@@ -196,14 +201,16 @@ acpi_cpu_probe(device_t dev)
return (ENXIO);
handle = acpi_get_handle(dev);
+DEBUG_KNOB(1);
if (cpu_softc == NULL)
cpu_softc = kmalloc(sizeof(struct acpi_cpu_softc *) *
SMP_MAXCPU, M_TEMP /* XXX */, M_INTWAIT | M_ZERO);
-
+DEBUG_KNOB(2);
/* Get our Processor object. */
buf.Pointer = NULL;
buf.Length = ACPI_ALLOCATE_BUFFER;
status = AcpiEvaluateObject(handle, NULL, NULL, &buf);
+DEBUG_KNOB(3);
if (ACPI_FAILURE(status)) {
device_printf(dev, "probe failed to get Processor obj - %s\n",
AcpiFormatException(status));
@@ -221,18 +228,25 @@ acpi_cpu_probe(device_t dev)
* ProcId as a key, however, some boxes do not have the same values
* in their Processor object as the ProcId values in the MADT.
*/
- acpi_id = obj->Processor.ProcId;
+ acpi_id = proc_id = obj->Processor.ProcId;
AcpiOsFree(obj);
- if (acpi_pcpu_get_id(device_get_unit(dev), &acpi_id, &cpu_id) != 0)
+DEBUG_KNOB(4);
+ if (acpi_pcpu_get_id(device_get_unit(dev), &acpi_id, &cpu_id) != 0) {
+ if (bootverbose)
+ device_printf(dev, "ProcId %d: can't find cpu_id\n", proc_id);
return (ENXIO);
-
+ }
+ if (bootverbose) {
+ device_printf(dev, "ProcId %d: acpi_id=%d, cpu_id=%d\n", proc_id,
+ acpi_id, cpu_id);
+ }
/*
* Check if we already probed this processor. We scan the bus twice
* so it's possible we've already seen this one.
*/
if (cpu_softc[cpu_id] != NULL)
return (ENXIO);
-
+DEBUG_KNOB(5);
/* Get a count of Cx states for our device string. */
cx_count = 0;
buf.Pointer = NULL;
@@ -263,6 +277,7 @@ acpi_cpu_probe(device_t dev)
return (0);
}
+#undef DEBUG_KNOB
static int
acpi_cpu_attach(device_t dev)
[
Date Prev][
Date Next]
[
Thread Prev][
Thread Next]
[
Date Index][
Thread Index]