Implement VT-d capability detection on chipsets that have multiple

translation units with differing capabilities

From the author via Bugzilla:
---
When an attempt is made to passthrough a PCI device to a bhyve VM
(causing initialisation of IOMMU) on certain Intel chipsets using
VT-d the PCI bus stops working entirely. This issue occurs on the
E3-1275 v5 processor on C236 chipset and has also been encountered
by others on the forums with different hardware in the Skylake
series.

The chipset has two VT-d translation units. The issue is caused by
an attempt to use the VT-d device-IOTLB capability that is
supported by only the first unit for devices attached to the
second unit which lacks that capability. Only the capabilities of
the first unit are checked and are assumed to be the same for all
units.

Attached is a patch to rectify this issue by determining which
unit is responsible for the device being added to a domain and
then checking that unit's device-IOTLB capability. In addition to
this a few fixes have been made to other instances where the first
unit's capabilities are assumed for all units for domains they
share. In these cases a mutual set of capabilities is determined.
The patch should hopefully fix any bugs for current/future
hardware with multiple translation units supporting different
capabilities.

A description is on the forums at
https://forums.freebsd.org/threads/pci-passthrough-bhyve-usb-xhci.65235
The thread includes observations by other users of the bug
occurring, and description as well as confirmation of the fix.
I'd also like to thank Ordoban for their help.

---
Personally tested on a Skylake laptop, Skylake Xeon server, and
a Xeon-D-1541, passing through XHCI and NVMe functions.  Passthru
is hit-or-miss to the point of being unusable without this
patch.

PR: 229852
Submitted by: callum@aitchison.org
MFC after: 1 week

sys/amd64/vmm/intel/vtd.c

@@ -51,6 +51,8 @@ __FBSDID("$FreeBSD$");
  * Architecture Spec, September 2008.
  */
 
+#define VTD_DRHD_INCLUDE_PCI_ALL(Flags)  (((Flags) >> 0) & 0x1)
+
 /* Section 10.4 "Register Descriptions" */
 struct vtdmap {
 	volatile uint32_t	version;
@@ -116,10 +118,11 @@ struct domain {
 static SLIST_HEAD(, domain) domhead;
 
 #define	DRHD_MAX_UNITS	8
-static int		drhd_num;
+static ACPI_DMAR_HARDWARE_UNIT	*drhds[DRHD_MAX_UNITS];
-static struct vtdmap	*vtdmaps[DRHD_MAX_UNITS];
+static int			drhd_num;
-static int		max_domains;
+static struct vtdmap		*vtdmaps[DRHD_MAX_UNITS];
-typedef int		(*drhd_ident_func_t)(void);
+static int			max_domains;
+typedef int			(*drhd_ident_func_t)(void);
 
 static uint64_t root_table[PAGE_SIZE / sizeof(uint64_t)] __aligned(4096);
 static uint64_t ctx_tables[256][PAGE_SIZE / sizeof(uint64_t)] __aligned(4096);
@@ -175,6 +178,69 @@ domain_id(void)
 	return (id);
 }
 
+static struct vtdmap *
+vtd_device_scope(uint16_t rid)
+{
+	int i, remaining, pathremaining;
+	char *end, *pathend;
+	struct vtdmap *vtdmap;
+	ACPI_DMAR_HARDWARE_UNIT *drhd;
+	ACPI_DMAR_DEVICE_SCOPE *device_scope;
+	ACPI_DMAR_PCI_PATH *path;
+
+	for (i = 0; i < drhd_num; i++) {
+		drhd = drhds[i];
+
+		if (VTD_DRHD_INCLUDE_PCI_ALL(drhd->Flags)) {
+			/*
+			 * From Intel VT-d arch spec, version 3.0:
+			 * If a DRHD structure with INCLUDE_PCI_ALL flag Set is reported
+			 * for a Segment, it must be enumerated by BIOS after all other
+			 * DRHD structures for the same Segment.
+			 */
+			vtdmap = vtdmaps[i];
+			return(vtdmap);
+		}
+
+		end = (char *)drhd + drhd->Header.Length;
+		remaining = drhd->Header.Length - sizeof(ACPI_DMAR_HARDWARE_UNIT);
+		while (remaining > sizeof(ACPI_DMAR_DEVICE_SCOPE)) {
+			device_scope = (ACPI_DMAR_DEVICE_SCOPE *)(end - remaining);
+			remaining -= device_scope->Length;
+
+			switch (device_scope->EntryType){
+				/* 0x01 and 0x02 are PCI device entries */
+				case 0x01:
+				case 0x02:
+					break;
+				default:
+					continue;
+			}
+
+			if (PCI_RID2BUS(rid) != device_scope->Bus)
+				continue;
+
+			pathend = (char *)device_scope + device_scope->Length;
+			pathremaining = device_scope->Length - sizeof(ACPI_DMAR_DEVICE_SCOPE);
+			while (pathremaining >= sizeof(ACPI_DMAR_PCI_PATH)) {
+				path = (ACPI_DMAR_PCI_PATH *)(pathend - pathremaining);
+				pathremaining -= sizeof(ACPI_DMAR_PCI_PATH);
+
+				if (PCI_RID2SLOT(rid) != path->Device)
+					continue;
+				if (PCI_RID2FUNC(rid) != path->Function)
+					continue;
+
+				vtdmap = vtdmaps[i];
+				return (vtdmap);
+			}
+		}
+	}
+
+	/* No matching scope */
+	return (NULL);
+}
+
 static void
 vtd_wbflush(struct vtdmap *vtdmap)
 {
@@ -240,7 +306,7 @@ vtd_translation_disable(struct vtdmap *vtdmap)
 static int
 vtd_init(void)
 {
-	int i, units, remaining;
+	int i, units, remaining, tmp;
 	struct vtdmap *vtdmap;
 	vm_paddr_t ctx_paddr;
 	char *end, envname[32];
@@ -291,8 +357,9 @@ vtd_init(void)
 			break;
 
 		drhd = (ACPI_DMAR_HARDWARE_UNIT *)hdr;
-		vtdmaps[units++] = (struct vtdmap *)PHYS_TO_DMAP(drhd->Address);
+		drhds[units] = drhd;
-		if (units >= DRHD_MAX_UNITS)
+		vtdmaps[units] = (struct vtdmap *)PHYS_TO_DMAP(drhd->Address);
+		if (++units >= DRHD_MAX_UNITS)
 			break;
 		remaining -= hdr->Length;
 	}
@@ -302,12 +369,18 @@ vtd_init(void)
 
 skip_dmar:
 	drhd_num = units;
-	vtdmap = vtdmaps[0];
 
-	if (VTD_CAP_CM(vtdmap->cap) != 0)
+	max_domains = 64 * 1024; /* maximum valid value */
-		panic("vtd_init: invalid caching mode");
+	for (i = 0; i < drhd_num; i++){
+		vtdmap = vtdmaps[i];
 
-	max_domains = vtd_max_domains(vtdmap);
+		if (VTD_CAP_CM(vtdmap->cap) != 0)
+			panic("vtd_init: invalid caching mode");
+
+		/* take most compatible (minimum) value */
+		if ((tmp = vtd_max_domains(vtdmap)) < max_domains)
+			max_domains = tmp;
+	}
 
 	/*
 	 * Set up the root-table to point to the context-entry tables
@@ -373,7 +446,6 @@ vtd_add_device(void *arg, uint16_t rid)
 	struct vtdmap *vtdmap;
 	uint8_t bus;
 
-	vtdmap = vtdmaps[0];
 	bus = PCI_RID2BUS(rid);
 	ctxp = ctx_tables[bus];
 	pt_paddr = vtophys(dom->ptp);
@@ -385,6 +457,10 @@ vtd_add_device(void *arg, uint16_t rid)
 		      (uint16_t)(ctxp[idx + 1] >> 8));
 	}
 
+	if ((vtdmap = vtd_device_scope(rid)) == NULL)
+		panic("vtd_add_device: device %x is not in scope for "
+		      "any DMA remapping unit", rid);
+
 	/*
 	 * Order is important. The 'present' bit is set only after all fields
 	 * of the context pointer are initialized.
@@ -568,8 +644,6 @@ vtd_create_domain(vm_paddr_t maxaddr)
 	if (drhd_num <= 0)
 		panic("vtd_create_domain: no dma remapping hardware available");
 
-	vtdmap = vtdmaps[0];
-
 	/*
 	 * Calculate AGAW.
 	 * Section 3.4.2 "Adjusted Guest Address Width", Architecture Spec.
@@ -594,7 +668,14 @@ vtd_create_domain(vm_paddr_t maxaddr)
 	pt_levels = 2;
 	sagaw = 30;
 	addrwidth = 0;
-	tmp = VTD_CAP_SAGAW(vtdmap->cap);
+
+	tmp = ~0;
+	for (i = 0; i < drhd_num; i++) {
+		vtdmap = vtdmaps[i];
+		/* take most compatible value */
+		tmp &= VTD_CAP_SAGAW(vtdmap->cap);
+	}
+
 	for (i = 0; i < 5; i++) {
 		if ((tmp & (1 << i)) != 0 && sagaw >= agaw)
 			break;
@@ -606,8 +687,8 @@ vtd_create_domain(vm_paddr_t maxaddr)
 	}
 
 	if (i >= 5) {
-		panic("vtd_create_domain: SAGAW 0x%lx does not support AGAW %d",
+		panic("vtd_create_domain: SAGAW 0x%x does not support AGAW %d",
-		      VTD_CAP_SAGAW(vtdmap->cap), agaw);
+		      tmp, agaw);
 	}
 
 	dom = malloc(sizeof(struct domain), M_VTD, M_ZERO | M_WAITOK);
@@ -634,7 +715,12 @@ vtd_create_domain(vm_paddr_t maxaddr)
 	 * There is not any code to deal with the demotion at the moment
 	 * so we disable superpage mappings altogether.
 	 */
-	dom->spsmask = VTD_CAP_SPS(vtdmap->cap);
+	dom->spsmask = ~0;
+	for (i = 0; i < drhd_num; i++) {
+		vtdmap = vtdmaps[i];
+		/* take most compatible value */
+		dom->spsmask &= VTD_CAP_SPS(vtdmap->cap);
+	}
 #endif
 
 	SLIST_INSERT_HEAD(&domhead, dom, next);