6775011 bad trap page fault while starting dom0

8 files changed, 141 insertions, 45 deletions

diff --git a/usr/src/uts/i86pc/os/cpuid.c b/usr/src/uts/i86pc/os/cpuid.c
index a82d6d70e5..15969b2c02 100644
--- a/usr/src/uts/i86pc/os/cpuid.c
+++ b/usr/src/uts/i86pc/os/cpuid.c
@@ -114,13 +114,9 @@ uint_t pentiumpro_bug4064495;
 
 uint_t enable486;
 /*
- * This is set if Solaris is booted in a fully virtualized mode:
- *	- as HVM guest under xVM
- *	- as guest under VMware
- * check_for_hvm() has the logic to detect these 2 cases.
- * This is not applicable if Solaris is booted as a para virtual guest.
+ * This is set to platform type Solaris is running on.
  */
-int platform_is_virt = 0;
+static int platform_type = HW_NATIVE;
 
 /*
  * monitor/mwait info.
@@ -429,12 +425,11 @@ cpuid_free_space(cpu_t *cpu)
 #if !defined(__xpv)
 
 static void
-check_for_hvm()
+determine_platform()
 {
 	struct cpuid_regs cp;
 	char *xen_str;
 	uint32_t xen_signature[4];
-	extern int xpv_is_hvm;
 
 	/*
 	 * In a fully virtualized domain, Xen's pseudo-cpuid function
@@ -449,13 +444,39 @@ check_for_hvm()
 	xen_signature[2] = cp.cp_edx;
 	xen_signature[3] = 0;
 	xen_str = (char *)xen_signature;
-	if (strcmp("XenVMMXenVMM", xen_str) == 0 && cp.cp_eax <= 0x40000002)
-		xpv_is_hvm = 1;
+	if (strcmp("XenVMMXenVMM", xen_str) == 0 && cp.cp_eax <= 0x40000002) {
+		platform_type = HW_XEN_HVM;
+	} else if (vmware_platform()) { /* running under vmware hypervisor? */
+		platform_type = HW_VMWARE;
+	}
+}
+
+int
+get_hwenv(void)
+{
+	return (platform_type);
+}
+
+int
+is_controldom(void)
+{
+	return (0);
+}
+
+#else
+
+int
+get_hwenv(void)
+{
+	return (HW_XEN_PV);
+}
 
-	/* could we be running under vmware hypervisor */
-	if (xpv_is_hvm || vmware_platform())
-		platform_is_virt = 1;
+int
+is_controldom(void)
+{
+	return (DOMAIN_IS_INITDOMAIN(xen_info));
 }
+
 #endif	/* __xpv */
 
 uint_t
@@ -1200,7 +1221,7 @@ cpuid_pass1(cpu_t *cpu)
 
 pass1_done:
 #if !defined(__xpv)
-	check_for_hvm();
+	determine_platform();
 #endif
 	cpi->cpi_pass = 1;
 	return (feature);
diff --git a/usr/src/uts/i86pc/os/microcode.c b/usr/src/uts/i86pc/os/microcode.c
index 8cf5d99d24..c69c8c0500 100644
--- a/usr/src/uts/i86pc/os/microcode.c
+++ b/usr/src/uts/i86pc/os/microcode.c
@@ -203,32 +203,22 @@ ucode_free(processorid_t id, void* buf, size_t size)
 static int
 ucode_capable_amd(cpu_t *cp)
 {
-#ifndef	__xpv
-	extern int xpv_is_hvm;
-	if (xpv_is_hvm) {
-		return (0);
-	}
-#else
-	if (!DOMAIN_IS_INITDOMAIN(xen_info)) {
+	int hwenv = get_hwenv();
+
+	if (hwenv == HW_XEN_HVM || (hwenv == HW_XEN_PV && !is_controldom())) {
 		return (0);
 	}
-#endif
 	return (cpuid_getfamily(cp) >= 0x10);
 }
 
 static int
 ucode_capable_intel(cpu_t *cp)
 {
-#ifndef	__xpv
-	extern int xpv_is_hvm;
-	if (xpv_is_hvm) {
-		return (0);
-	}
-#else
-	if (!DOMAIN_IS_INITDOMAIN(xen_info)) {
+	int hwenv = get_hwenv();
+
+	if (hwenv == HW_XEN_HVM || (hwenv == HW_XEN_PV && !is_controldom())) {
 		return (0);
 	}
-#endif
 	return (cpuid_getfamily(cp) >= 6);
 }
 
diff --git a/usr/src/uts/i86pc/os/mlsetup.c b/usr/src/uts/i86pc/os/mlsetup.c
index 001bd0537f..c3aa711f13 100644
--- a/usr/src/uts/i86pc/os/mlsetup.c
+++ b/usr/src/uts/i86pc/os/mlsetup.c
@@ -102,9 +102,6 @@ mlsetup(struct regs *rp)
 	extern struct classfuncs sys_classfuncs;
 	extern disp_t cpu0_disp;
 	extern char t0stack[];
-#if !defined(__xpv)
-	extern int xpv_is_hvm;
-#endif
 
 	ASSERT_STACK_ALIGNED();
 
@@ -180,7 +177,7 @@ mlsetup(struct regs *rp)
 	 * The Xen hypervisor does not correctly report whether rdtscp is
 	 * supported or not, so we must assume that it is not.
 	 */
-	if (xpv_is_hvm == 0 && (x86_feature & X86_TSCP))
+	if (get_hwenv() != HW_XEN_HVM && (x86_feature & X86_TSCP))
 		patch_tsc_read(X86_HAVE_TSCP);
 	else if (cpuid_getvendor(CPU) == X86_VENDOR_AMD &&
 	    cpuid_getfamily(CPU) <= 0xf && (x86_feature & X86_SSE2) != 0)
diff --git a/usr/src/uts/i86pc/os/startup.c b/usr/src/uts/i86pc/os/startup.c
index 6083fb21cd..119bfabf2f 100644
--- a/usr/src/uts/i86pc/os/startup.c
+++ b/usr/src/uts/i86pc/os/startup.c
@@ -541,11 +541,6 @@ static int lower_pages_count = 0;
 struct system_hardware system_hardware;
 
 /*
- * Is this Solaris instance running in a fully virtualized xVM domain?
- */
-int xpv_is_hvm = 0;
-
-/*
  * Enable some debugging messages concerning memory usage...
  */
 static void
@@ -1431,7 +1426,7 @@ startup_modules(void)
 	 */
 	microfind();
 
-	if (xpv_is_hvm)
+	if (get_hwenv() == HW_XEN_HVM)
 		update_default_path();
 #endif
 
@@ -1542,7 +1537,7 @@ startup_modules(void)
 	 * Initialize a handle for the boot cpu - others will initialize
 	 * as they startup.  Do not do this if we know we are in an HVM domU.
 	 */
-	if (!xpv_is_hvm &&
+	if ((get_hwenv() != HW_XEN_HVM) &&
 	    (hdl = cmi_init(CMI_HDL_NATIVE, cmi_ntv_hwchipid(CPU),
 	    cmi_ntv_hwcoreid(CPU), cmi_ntv_hwstrandid(CPU))) != NULL &&
 	    (x86_feature & X86_MCA))
@@ -2155,7 +2150,7 @@ post_startup(void)
 		 * Startup the memory scrubber.
 		 * XXPV	This should be running somewhere ..
 		 */
-		if (!xpv_is_hvm)
+		if (get_hwenv() != HW_XEN_HVM)
 			memscrub_init();
 #endif
 	}
diff --git a/usr/src/uts/i86pc/os/timestamp.c b/usr/src/uts/i86pc/os/timestamp.c
index e93f9de2c8..02b4cb529c 100644
--- a/usr/src/uts/i86pc/os/timestamp.c
+++ b/usr/src/uts/i86pc/os/timestamp.c
@@ -134,7 +134,6 @@ static volatile int tsc_sync_go;
 }
 
 int tsc_master_slave_sync_needed = 1;
-extern int platform_is_virt;
 
 static int	tsc_max_delta;
 static hrtime_t tsc_sync_tick_delta[NCPU];
@@ -415,8 +414,11 @@ tsc_sync_master(processorid_t slave)
 	hrtime_t write_time, x, mtsc_after, tdelta;
 	tsc_sync_t *tsc = tscp;
 	int cnt;
+	int hwtype;
 
-	if (!tsc_master_slave_sync_needed || platform_is_virt)
+	hwtype = get_hwenv();
+	if (!tsc_master_slave_sync_needed || hwtype == HW_XEN_HVM ||
+	    hwtype == HW_VMWARE)
 		return;
 
 	flags = clear_int_flag();
@@ -492,8 +494,11 @@ tsc_sync_slave(void)
 	hrtime_t s1;
 	tsc_sync_t *tsc = tscp;
 	int cnt;
+	int hwtype;
 
-	if (!tsc_master_slave_sync_needed || platform_is_virt)
+	hwtype = get_hwenv();
+	if (!tsc_master_slave_sync_needed || hwtype == HW_XEN_HVM ||
+	    hwtype == HW_VMWARE)
 		return;
 
 	flags = clear_int_flag();
diff --git a/usr/src/uts/intel/io/acpica/acpica.c b/usr/src/uts/intel/io/acpica/acpica.c
index 283e06a164..5889ee845a 100644
--- a/usr/src/uts/intel/io/acpica/acpica.c
+++ b/usr/src/uts/intel/io/acpica/acpica.c
@@ -36,6 +36,7 @@
 #include <sys/sunddi.h>
 #include <sys/esunddi.h>
 #include <sys/kstat.h>
+#include <sys/x86_archext.h>
 
 #include <sys/acpi/acpi.h>
 #include <sys/acpica.h>
@@ -370,6 +371,7 @@ acpica_process_user_options()
 int
 acpica_init()
 {
+	extern void acpica_find_ioapics(void);
 	ACPI_STATUS status;
 
 	/*
@@ -406,6 +408,13 @@ acpica_init()
 		acpica_ec_init();
 
 		acpica_init_state = ACPICA_INITIALIZED;
+		/*
+		 * If we are running on the Xen hypervisor as dom0 we need to
+		 * find the ioapics so we can prevent ACPI from trying to
+		 * access them.
+		 */
+		if (get_hwenv() == HW_XEN_PV && is_controldom())
+			acpica_find_ioapics();
 		acpica_init_kstats();
 error:
 		if (acpica_init_state != ACPICA_INITIALIZED) {
diff --git a/usr/src/uts/intel/io/acpica/osl.c b/usr/src/uts/intel/io/acpica/osl.c
index b292fcfa62..b0991d35f7 100644
--- a/usr/src/uts/intel/io/acpica/osl.c
+++ b/usr/src/uts/intel/io/acpica/osl.c
@@ -31,6 +31,7 @@
 #include <sys/kmem.h>
 #include <sys/psm.h>
 #include <sys/pci_cfgspace.h>
+#include <sys/apic.h>
 #include <sys/ddi.h>
 #include <sys/sunddi.h>
 #include <sys/sunndi.h>
@@ -518,9 +519,71 @@ AcpiOsFree(void *Memory)
 	kmem_free(tmp_ptr, size);
 }
 
+static int napics_found;	/* number of ioapic addresses in array */
+static ACPI_PHYSICAL_ADDRESS ioapic_paddr[MAX_IO_APIC];
+static ACPI_TABLE_MADT *acpi_mapic_dtp = NULL;
+static void *dummy_ioapicadr;
+
+void
+acpica_find_ioapics(void)
+{
+	int			madt_seen, madt_size;
+	ACPI_SUBTABLE_HEADER		*ap;
+	ACPI_MADT_IO_APIC		*mia;
+
+	if (acpi_mapic_dtp != NULL)
+		return;	/* already parsed table */
+	if (AcpiGetTable(ACPI_SIG_MADT, 1,
+	    (ACPI_TABLE_HEADER **) &acpi_mapic_dtp) != AE_OK)
+		return;
+
+	napics_found = 0;
+
+	/*
+	 * Search the MADT for ioapics
+	 */
+	ap = (ACPI_SUBTABLE_HEADER *) (acpi_mapic_dtp + 1);
+	madt_size = acpi_mapic_dtp->Header.Length;
+	madt_seen = sizeof (*acpi_mapic_dtp);
+
+	while (madt_seen < madt_size) {
+
+		switch (ap->Type) {
+		case ACPI_MADT_TYPE_IO_APIC:
+			mia = (ACPI_MADT_IO_APIC *) ap;
+			if (napics_found < MAX_IO_APIC) {
+				ioapic_paddr[napics_found++] =
+				    (ACPI_PHYSICAL_ADDRESS)
+				    (mia->Address & PAGEMASK);
+			}
+			break;
+
+		default:
+			break;
+		}
+
+		/* advance to next entry */
+		madt_seen += ap->Length;
+		ap = (ACPI_SUBTABLE_HEADER *)(((char *)ap) + ap->Length);
+	}
+	if (dummy_ioapicadr == NULL)
+		dummy_ioapicadr = kmem_zalloc(PAGESIZE, KM_SLEEP);
+}
+
+
 void *
 AcpiOsMapMemory(ACPI_PHYSICAL_ADDRESS PhysicalAddress, ACPI_SIZE Size)
 {
+	int	i;
+
+	/*
+	 * If the iopaic address table is populated, check if trying
+	 * to access an ioapic.  Instead, return a pointer to a dummy ioapic.
+	 */
+	for (i = 0; i < napics_found; i++) {
+		if ((PhysicalAddress & PAGEMASK) == ioapic_paddr[i])
+			return (dummy_ioapicadr);
+	}
 	/* FUTUREWORK: test PhysicalAddress for > 32 bits */
 	return (psm_map_new((paddr_t)PhysicalAddress,
 	    (size_t)Size, PSM_PROT_WRITE | PSM_PROT_READ));
@@ -529,6 +592,11 @@ AcpiOsMapMemory(ACPI_PHYSICAL_ADDRESS PhysicalAddress, ACPI_SIZE Size)
 void
 AcpiOsUnmapMemory(void *LogicalAddress, ACPI_SIZE Size)
 {
+	/*
+	 * Check if trying to unmap dummy ioapic address.
+	 */
+	if (LogicalAddress == dummy_ioapicadr)
+		return;
 
 	psm_unmap((caddr_t)LogicalAddress, (size_t)Size);
 }
diff --git a/usr/src/uts/intel/sys/x86_archext.h b/usr/src/uts/intel/sys/x86_archext.h
index 61a93b7556..8a84d0e251 100644
--- a/usr/src/uts/intel/sys/x86_archext.h
+++ b/usr/src/uts/intel/sys/x86_archext.h
@@ -672,6 +672,17 @@ extern int opteron_workaround_6323525;
 extern void patch_workaround_6323525(void);
 #endif
 
+extern int get_hwenv(void);
+extern int is_controldom(void);
+
+/*
+ * Defined hardware environments
+ */
+#define	HW_NATIVE	0x00	/* Running on bare metal */
+#define	HW_XEN_PV	0x01	/* Running on Xen Hypervisor paravirutualized */
+#define	HW_XEN_HVM	0x02	/* Running on Xen hypervisor HVM */
+#define	HW_VMWARE	0x03	/* Running on VMware hypervisor */
+
 #endif	/* _KERNEL */
 
 #endif