Failure to get past a PCI bridge
John Baldwin
jhb at freebsd.org
Fri Jun 5 14:35:13 UTC 2009
On Friday 05 June 2009 5:17:25 am Josef Moellers wrote:
> Difficult, since I can't boot properly.
> However, I have managed to get the dsdt using a SuSE Linux and have run
> that through acpidump -d on a 7.2 running on a XEN virtual machine.
> Here's the result.
Hmm, your BIOS is certainly hosed. First, it does have separate processor
objects:
Scope (\_PR)
{
Processor (PR00, 0x00, 0x00000000, 0x00) {}
Processor (PR01, 0x08, 0x00000000, 0x00) {}
Processor (PR02, 0x01, 0x00000000, 0x00) {}
Processor (PR03, 0x09, 0x00000000, 0x00) {}
Processor (PR04, 0x02, 0x00000000, 0x00) {}
Processor (PR05, 0x0A, 0x00000000, 0x00) {}
Processor (PR06, 0x03, 0x00000000, 0x00) {}
Processor (PR07, 0x0B, 0x00000000, 0x00) {}
Processor (PR10, 0x04, 0x00000000, 0x00) {}
Processor (PR11, 0x0C, 0x00000000, 0x00) {}
Processor (PR12, 0x05, 0x00000000, 0x00) {}
Processor (PR13, 0x0D, 0x00000000, 0x00) {}
Processor (PR14, 0x06, 0x00000000, 0x00) {}
Processor (PR15, 0x0E, 0x00000000, 0x00) {}
Processor (PR16, 0x07, 0x00000000, 0x00) {}
Processor (PR17, 0x0F, 0x00000000, 0x00) {}
}
However, \_SB_.CPU0 and \_SB_.CPU1 are listed as Host-PCI bridge devices, not
CPUs (CPUs are enumerated in ACPI as Processor() objects, not Device()'s, and
certainly not with the _HID == Host-PCI bridge):
Device (CPU0)
{
Name (_HID, EisaId ("PNP0A03"))
Name (_UID, 0x01)
Name (_BBN, 0xFF)
...
Device (CPU1)
{
Name (_HID, EisaId ("PNP0A03"))
Name (_UID, 0x02)
Name (_BBN, 0xFE)
...
Device (PCI0)
{
Name (_HID, EisaId ("PNP0A08"))
Name (_CID, 0x030AD041)
Name (_UID, 0x00)
Name (_BBN, 0x00)
Method (_INI, 0, NotSerialized)
{
\OSCK ()
}
Hmm, it does have a valid _BBN value for your actual PCI bus. Not sure why
it is using a bus number of 2. Ahh, ok. Hmm, so some machines with multipe
busses would set _BBN to 0 for all busses, so we don't trust a _BBN of 0 for
non-pcib0. However, I think what we want to do is not trust a _BBN of 0 if
we already have a PCI bus 0 (in your case, pcib0 has PCI bus 255, pcib1 has
PCI bus 254, and pcib2 has PCI bus 0). Try this:
--- //depot/vendor/freebsd/src/sys/dev/acpica/acpi_pcib_acpi.c 2009/02/05 18:45:20
+++ //depot/user/jhb/acpipci/dev/acpica/acpi_pcib_acpi.c 2009/06/05 14:28:27
@@ -146,6 +146,7 @@
{
struct acpi_hpcib_softc *sc;
ACPI_STATUS status;
+ static int bus0_seen = 0;
u_int addr, slot, func, busok;
uint8_t busno;
@@ -156,6 +157,21 @@
sc->ap_handle = acpi_get_handle(dev);
/*
+ * Get our segment number by evaluating _SEG
+ * It's OK for this to not exist.
+ */
+ status = acpi_GetInteger(sc->ap_handle, "_SEG", &sc->ap_segment);
+ if (ACPI_FAILURE(status)) {
+ if (status != AE_NOT_FOUND) {
+ device_printf(dev, "could not evaluate _SEG - %s\n",
+ AcpiFormatException(status));
+ return_VALUE (ENXIO);
+ }
+ /* If it's not found, assume 0. */
+ sc->ap_segment = 0;
+ }
+
+ /*
* Get our base bus number by evaluating _BBN.
* If this doesn't work, we assume we're bus number 0.
*
@@ -169,8 +185,10 @@
* XXX invoke _REG on this for the PCI config space address space?
* XXX It seems many BIOS's with multiple Host-PCI bridges do not set
* _BBN correctly. They set _BBN to zero for all bridges. Thus,
- * if _BBN is zero and pcib0 already exists, we try to read our
+ * if _BBN is zero and PCI bus 0 already exists, we try to read our
* bus number from the configuration registers at address _ADR.
+ * We only do this for domain/segment 0 in the hopes that this is
+ * only needed for old single-domain machines.
*/
status = acpi_GetInteger(sc->ap_handle, "_BBN", &sc->ap_bus);
if (ACPI_FAILURE(status)) {
@@ -185,11 +203,11 @@
}
/*
- * If the bus is zero and pcib0 already exists, read the bus number
- * via PCI config space.
+ * If this is segment 0, the bus is zero, and PCI bus 0 already
+ * exists, read the bus number via PCI config space.
*/
busok = 1;
- if (sc->ap_bus == 0 && devclass_get_device(pcib_devclass, 0) != dev) {
+ if (sc->ap_segment == 0 && sc->ap_bus == 0 && bus0_seen) {
busok = 0;
status = acpi_GetInteger(sc->ap_handle, "_ADR", &addr);
if (ACPI_FAILURE(status)) {
@@ -226,20 +244,9 @@
device_printf(dev, "trying bus number %d\n", sc->ap_bus);
}
- /*
- * Get our segment number by evaluating _SEG
- * It's OK for this to not exist.
- */
- status = acpi_GetInteger(sc->ap_handle, "_SEG", &sc->ap_segment);
- if (ACPI_FAILURE(status)) {
- if (status != AE_NOT_FOUND) {
- device_printf(dev, "could not evaluate _SEG - %s\n",
- AcpiFormatException(status));
- return_VALUE (ENXIO);
- }
- /* If it's not found, assume 0. */
- sc->ap_segment = 0;
- }
+ /* If this is bus 0 on segment 0, note that it has been seen already. */
+ if (sc->ap_segment == 0 && sc->ap_bus == 0)
+ bus0_seen = 1;
return (acpi_pcib_attach(dev, &sc->ap_prt, sc->ap_bus));
}
--
John Baldwin
More information about the freebsd-acpi
mailing list