Don't allow MPtable generation if there are multiple PCI hierarchies. This is

because there isn't a standard way to relay this information to the guest OS.

Add a command line option "-Y" to bhyve(8) to inhibit MPtable generation.

If the virtual machine is using PCI devices on buses other than 0 then it can
still use ACPI tables to convey this information to the guest.

Discussed with:	grehan@

5 files changed, 39 insertions, 5 deletions

diff --git a/usr.sbin/bhyve/bhyve.8 b/usr.sbin/bhyve/bhyve.8
index 88e72142b2ca..98062b7805f5 100644
--- a/usr.sbin/bhyve/bhyve.8
+++ b/usr.sbin/bhyve/bhyve.8
@@ -236,6 +236,8 @@ This is intended for debug purposes.
 Ignore accesses to unimplemented Model Specific Registers (MSRs). This is intended for debug purposes.
 .It Fl x
 The guest's local APIC is configured in x2APIC mode.
+.It Fl Y
+Disable MPtable generation.
 .It Fl h
 Print help message and exit.
 .It Ar vmname
diff --git a/usr.sbin/bhyve/bhyverun.c b/usr.sbin/bhyve/bhyverun.c
index b54e7148d943..ede7ab56169d 100644
--- a/usr.sbin/bhyve/bhyverun.c
+++ b/usr.sbin/bhyve/bhyverun.c
@@ -145,6 +145,7 @@ usage(int code)
 		"       -m: memory size in MB\n"
 		"       -w: ignore unimplemented MSRs\n"
 		"       -x: local apic is in x2APIC mode\n"
+		"       -Y: disable MPtable generation\n"
 		"       -U: uuid\n",
 		progname, (int)strlen(progname), "");
 
@@ -616,7 +617,7 @@ int
 main(int argc, char *argv[])
 {
 	int c, error, gdb_port, err, bvmcons;
-	int max_vcpus;
+	int max_vcpus, mptgen;
 	struct vmctx *ctx;
 	uint64_t rip;
 	size_t memsize;
@@ -626,8 +627,9 @@ main(int argc, char *argv[])
 	gdb_port = 0;
 	guest_ncpus = 1;
 	memsize = 256 * MB;
+	mptgen = 1;
 
-	while ((c = getopt(argc, argv, "abehwxAHIPWp:g:c:s:m:l:U:")) != -1) {
+	while ((c = getopt(argc, argv, "abehwxAHIPWYp:g:c:s:m:l:U:")) != -1) {
 		switch (c) {
 		case 'a':
 			x2apic_mode = 0;
@@ -693,6 +695,9 @@ main(int argc, char *argv[])
 		case 'x':
 			x2apic_mode = 1;
 			break;
+		case 'Y':
+			mptgen = 0;
+			break;
 		case 'h':
 			usage(0);			
 		default:
@@ -752,7 +757,11 @@ main(int argc, char *argv[])
 	/*
 	 * build the guest tables, MP etc.
 	 */
-	mptable_build(ctx, guest_ncpus);
+	if (mptgen) {
+		error = mptable_build(ctx, guest_ncpus);
+		if (error)
+			exit(1);
+	}
 
 	error = smbios_build(ctx);
 	assert(error == 0);
diff --git a/usr.sbin/bhyve/mptbl.c b/usr.sbin/bhyve/mptbl.c
index e8958359adb6..4c2167e98c85 100644
--- a/usr.sbin/bhyve/mptbl.c
+++ b/usr.sbin/bhyve/mptbl.c
@@ -303,16 +303,31 @@ mptable_build(struct vmctx *ctx, int ncpu)
 	proc_entry_ptr		mpep;
 	mpfps_t			mpfp;
 	int_entry_ptr		mpie;
-	int			ioints;
+	int			ioints, bus;
 	char 			*curraddr;
 	char 			*startaddr;
 
 	startaddr = paddr_guest2host(ctx, MPTABLE_BASE, MPTABLE_MAX_LENGTH);
 	if (startaddr == NULL) {
-		printf("mptable requires mapped mem\n");
+		fprintf(stderr, "mptable requires mapped mem\n");
 		return (ENOMEM);
 	}
 
+	/*
+	 * There is no way to advertise multiple PCI hierarchies via MPtable
+	 * so require that there is no PCI hierarchy with a non-zero bus
+	 * number.
+	 */
+	for (bus = 1; bus <= PCI_BUSMAX; bus++) {
+		if (pci_bus_configured(bus)) {
+			fprintf(stderr, "MPtable is incompatible with "
+			    "multiple PCI hierarchies.\r\n");
+			fprintf(stderr, "MPtable generation can be disabled "
+			    "by passing the -Y option to bhyve(8).\r\n");
+			return (EINVAL);
+		}
+	}
+
 	curraddr = startaddr;
 	mpfp = (mpfps_t)curraddr;
 	mpt_build_mpfp(mpfp, MPTABLE_BASE);
diff --git a/usr.sbin/bhyve/pci_emul.c b/usr.sbin/bhyve/pci_emul.c
index 15f2fad53485..5b87da7f0604 100644
--- a/usr.sbin/bhyve/pci_emul.c
+++ b/usr.sbin/bhyve/pci_emul.c
@@ -1261,6 +1261,13 @@ pci_write_dsdt(void)
 }
 
 int
+pci_bus_configured(int bus)
+{
+	assert(bus >= 0 && bus < MAXBUSES);
+	return (pci_businfo[bus] != NULL);
+}
+
+int
 pci_msi_enabled(struct pci_devinst *pi)
 {
 	return (pi->pi_msi.enabled);
diff --git a/usr.sbin/bhyve/pci_emul.h b/usr.sbin/bhyve/pci_emul.h
index 7ad0a5d1f6b2..e1040a8b715a 100644
--- a/usr.sbin/bhyve/pci_emul.h
+++ b/usr.sbin/bhyve/pci_emul.h
@@ -233,6 +233,7 @@ uint64_t pci_emul_msix_tread(struct pci_devinst *pi, uint64_t offset, int size);
 int	pci_count_lintr(int bus);
 void	pci_walk_lintr(int bus, pci_lintr_cb cb, void *arg);
 void	pci_write_dsdt(void);
+int	pci_bus_configured(int bus);
 
 static __inline void 
 pci_set_cfgdata8(struct pci_devinst *pi, int offset, uint8_t val)