Respect the BIOS/firmware's notion of acceptable address ranges for PCI

resource allocation on x86 platforms:
- Add a new helper API that Host-PCI bridge drivers can use to restrict
  resource allocation requests to a set of address ranges for different
  resource types.
- For the ACPI Host-PCI bridge driver, use Producer address range resources
  in _CRS to enumerate valid address ranges for a given Host-PCI bridge.
  This can be disabled by including "hostres" in the debug.acpi.disabled
  tunable.
- For the MPTable Host-PCI bridge driver, use entries in the extended
  MPTable to determine the valid address ranges for a given Host-PCI
  bridge.  This required adding code to parse extended table entries.

Similar to the new PCI-PCI bridge driver, these changes are only enabled
if the NEW_PCIB kernel option is enabled (which is enabled by default on
amd64 and i386).

Approved by:	re (kib)

4 files changed, 350 insertions, 13 deletions

diff --git a/sys/dev/acpica/acpi_pcib_acpi.c b/sys/dev/acpica/acpi_pcib_acpi.c
index 73b2f70..d3d499e 100644
--- a/sys/dev/acpica/acpi_pcib_acpi.c
+++ b/sys/dev/acpica/acpi_pcib_acpi.c
@@ -32,6 +32,7 @@ __FBSDID("$FreeBSD$");
 #include <sys/param.h>
 #include <sys/bus.h>
 #include <sys/kernel.h>
+#include <sys/limits.h>
 #include <sys/malloc.h>
 #include <sys/module.h>
 #include <sys/rman.h>
@@ -62,6 +63,9 @@ struct acpi_hpcib_softc {
     int			ap_bus;		/* bios-assigned bus number */
 
     ACPI_BUFFER		ap_prt;		/* interrupt routing table */
+#ifdef NEW_PCIB
+    struct pcib_host_resources ap_host_res;
+#endif
 };
 
 static int		acpi_pcib_acpi_probe(device_t bus);
@@ -87,6 +91,11 @@ static struct resource *acpi_pcib_acpi_alloc_resource(device_t dev,
 			    device_t child, int type, int *rid,
 			    u_long start, u_long end, u_long count,
 			    u_int flags);
+#ifdef NEW_PCIB
+static int		acpi_pcib_acpi_adjust_resource(device_t dev,
+			    device_t child, int type, struct resource *r,
+			    u_long start, u_long end);
+#endif
 
 static device_method_t acpi_pcib_acpi_methods[] = {
     /* Device interface */
@@ -101,7 +110,11 @@ static device_method_t acpi_pcib_acpi_methods[] = {
     DEVMETHOD(bus_read_ivar,		acpi_pcib_read_ivar),
     DEVMETHOD(bus_write_ivar,		acpi_pcib_write_ivar),
     DEVMETHOD(bus_alloc_resource,	acpi_pcib_acpi_alloc_resource),
+#ifdef NEW_PCIB
+    DEVMETHOD(bus_adjust_resource,	acpi_pcib_acpi_adjust_resource),
+#else
     DEVMETHOD(bus_adjust_resource,	bus_generic_adjust_resource),
+#endif
     DEVMETHOD(bus_release_resource,	bus_generic_release_resource),
     DEVMETHOD(bus_activate_resource,	bus_generic_activate_resource),
     DEVMETHOD(bus_deactivate_resource,	bus_generic_deactivate_resource),
@@ -149,6 +162,120 @@ acpi_pcib_acpi_probe(device_t dev)
     return (0);
 }
 
+#ifdef NEW_PCIB
+static ACPI_STATUS
+acpi_pcib_producer_handler(ACPI_RESOURCE *res, void *context)
+{
+	struct acpi_hpcib_softc *sc;
+	UINT64 length, min, max;
+	u_int flags;
+	int error, type;
+
+	sc = context;
+	switch (res->Type) {
+	case ACPI_RESOURCE_TYPE_START_DEPENDENT:
+	case ACPI_RESOURCE_TYPE_END_DEPENDENT:
+		panic("host bridge has depenedent resources");
+	case ACPI_RESOURCE_TYPE_ADDRESS16:
+	case ACPI_RESOURCE_TYPE_ADDRESS32:
+	case ACPI_RESOURCE_TYPE_ADDRESS64:
+	case ACPI_RESOURCE_TYPE_EXTENDED_ADDRESS64:
+		if (res->Data.Address.ProducerConsumer != ACPI_PRODUCER)
+			break;
+		switch (res->Type) {
+		case ACPI_RESOURCE_TYPE_ADDRESS16:
+			min = res->Data.Address16.Minimum;
+			max = res->Data.Address16.Maximum;
+			length = res->Data.Address16.AddressLength;
+			break;
+		case ACPI_RESOURCE_TYPE_ADDRESS32:
+			min = res->Data.Address32.Minimum;
+			max = res->Data.Address32.Maximum;
+			length = res->Data.Address32.AddressLength;
+			break;
+		case ACPI_RESOURCE_TYPE_ADDRESS64:
+			min = res->Data.Address64.Minimum;
+			max = res->Data.Address64.Maximum;
+			length = res->Data.Address64.AddressLength;
+			break;
+		default:
+			KASSERT(res->Type ==
+			    ACPI_RESOURCE_TYPE_EXTENDED_ADDRESS64,
+			    ("should never happen"));
+			min = res->Data.ExtAddress64.Minimum;
+			max = res->Data.ExtAddress64.Maximum;
+			length = res->Data.ExtAddress64.AddressLength;
+			break;
+		}
+		if (length == 0 ||
+		    res->Data.Address.MinAddressFixed != ACPI_ADDRESS_FIXED ||
+		    res->Data.Address.MaxAddressFixed != ACPI_ADDRESS_FIXED)
+			break;
+		flags = 0;
+		switch (res->Data.Address.ResourceType) {
+		case ACPI_MEMORY_RANGE:
+			type = SYS_RES_MEMORY;
+			if (res->Type != ACPI_RESOURCE_TYPE_EXTENDED_ADDRESS64) {
+				if (res->Data.Address.Info.Mem.Caching ==
+				    ACPI_PREFETCHABLE_MEMORY)
+					flags |= RF_PREFETCHABLE;
+			} else {
+				/*
+				 * XXX: Parse prefetch flag out of
+				 * TypeSpecific.
+				 */
+			}
+			break;
+		case ACPI_IO_RANGE:
+			type = SYS_RES_IOPORT;
+			break;
+#ifdef PCI_RES_BUS
+		case ACPI_BUS_NUMBER_RANGE:
+			type = PCI_RES_BUS;
+			break;
+#endif
+		default:
+			return (AE_OK);
+		}
+
+		/* XXX: Not sure this is correct? */
+		if (res->Data.Address.Decode != ACPI_POS_DECODE) {
+			device_printf(sc->ap_dev,
+		    "Ignoring %d range (%#jx-%#jx) due to negative decode\n",
+			    type, (uintmax_t)min, (uintmax_t)max);
+			break;
+		}
+		if (min + length - 1 != max)
+			device_printf(sc->ap_dev,
+			    "Length mismatch for %d range: %jx vs %jx\n", type,
+			    (uintmax_t)max - min + 1, (uintmax_t)length);
+#ifdef __i386__
+		if (min > ULONG_MAX) {
+			device_printf(sc->ap_dev,
+			    "Ignoring %d range above 4GB (%#jx-%#jx)\n",
+			    type, (uintmax_t)min, (uintmax_t)max);
+			break;
+		}
+		if (max > ULONG_MAX) {
+			device_printf(sc->ap_dev,
+       		    "Truncating end of %d range above 4GB (%#jx-%#jx)\n",
+			    type, (uintmax_t)min, (uintmax_t)max);
+			max = ULONG_MAX;
+		}
+#endif
+		error = pcib_host_res_decodes(&sc->ap_host_res, type, min, max,
+		    flags);
+		if (error)
+			panic("Failed to manage %d range (%#jx-%#jx): %d",
+			    type, (uintmax_t)min, (uintmax_t)max, error);
+		break;
+	default:
+		break;
+	}
+	return (AE_OK);
+}
+#endif
+
 static int
 acpi_pcib_acpi_attach(device_t dev)
 {
@@ -179,6 +306,22 @@ acpi_pcib_acpi_attach(device_t dev)
 	sc->ap_segment = 0;
     }
 
+#ifdef NEW_PCIB
+    /*
+     * Determine which address ranges this bridge decodes and setup
+     * resource managers for those ranges.
+     */
+    if (pcib_host_res_init(sc->ap_dev, &sc->ap_host_res) != 0)
+	    panic("failed to init hostb resources");
+    if (!acpi_disabled("hostres")) {
+	status = AcpiWalkResources(sc->ap_handle, "_CRS",
+	    acpi_pcib_producer_handler, sc);
+	if (ACPI_FAILURE(status) && status != AE_NOT_FOUND)
+	    device_printf(sc->ap_dev, "failed to parse resources: %s\n",
+		AcpiFormatException(status));
+    }
+#endif
+
     /*
      * Get our base bus number by evaluating _BBN.
      * If this doesn't work, we assume we're bus number 0.
@@ -361,10 +504,33 @@ struct resource *
 acpi_pcib_acpi_alloc_resource(device_t dev, device_t child, int type, int *rid,
     u_long start, u_long end, u_long count, u_int flags)
 {
+#ifdef NEW_PCIB
+    struct acpi_hpcib_softc *sc;
+#endif
 
 #if defined(__i386__) || defined(__amd64__)
     start = hostb_alloc_start(type, start, end, count);
 #endif
+
+#ifdef NEW_PCIB
+    sc = device_get_softc(dev);
+    return (pcib_host_res_alloc(&sc->ap_host_res, child, type, rid, start, end,
+	count, flags));
+#else
     return (bus_generic_alloc_resource(dev, child, type, rid, start, end,
 	count, flags));
+#endif
+}
+
+#ifdef NEW_PCIB
+int
+acpi_pcib_acpi_adjust_resource(device_t dev, device_t child, int type,
+    struct resource *r, u_long start, u_long end)
+{
+	struct acpi_hpcib_softc *sc;
+
+	sc = device_get_softc(dev);
+	return (pcib_host_res_adjust(&sc->ap_host_res, child, type, r, start,
+	    end));
 }
+#endif
diff --git a/sys/dev/pci/pci_pci.c b/sys/dev/pci/pci_pci.c
index 5cbfb4d..fa14816 100644
--- a/sys/dev/pci/pci_pci.c
+++ b/sys/dev/pci/pci_pci.c
@@ -767,18 +767,6 @@ pcib_write_ivar(device_t dev, device_t child, int which, uintptr_t value)
 }
 
 #ifdef NEW_PCIB
-static const char *
-pcib_child_name(device_t child)
-{
-	static char buf[64];
-
-	if (device_get_nameunit(child) != NULL)
-		return (device_get_nameunit(child));
-	snprintf(buf, sizeof(buf), "pci%d:%d:%d:%d", pci_get_domain(child),
-	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
-	return (buf);
-}
-
 /*
  * Attempt to allocate a resource from the existing resources assigned
  * to a window.
diff --git a/sys/dev/pci/pci_subr.c b/sys/dev/pci/pci_subr.c
index 7f0c806..2d16896 100644
--- a/sys/dev/pci/pci_subr.c
+++ b/sys/dev/pci/pci_subr.c
@@ -33,9 +33,10 @@ __FBSDID("$FreeBSD$");
  * provide PCI domains.
  */
 
-#include <sys/types.h>
+#include <sys/param.h>
 #include <sys/bus.h>
 #include <sys/rman.h>
+#include <sys/systm.h>
 
 #include <dev/pci/pcireg.h>
 #include <dev/pci/pcivar.h>
@@ -128,3 +129,157 @@ host_pcib_get_busno(pci_read_config_fn read_config, int bus, int slot, int func,
 
 	return 1;
 }
+
+#ifdef NEW_PCIB
+/*
+ * Return a pointer to a pretty name for a PCI device.  If the device
+ * has a driver attached, the device's name is used, otherwise a name
+ * is generated from the device's PCI address.
+ */
+const char *
+pcib_child_name(device_t child)
+{
+	static char buf[64];
+
+	if (device_get_nameunit(child) != NULL)
+		return (device_get_nameunit(child));
+	snprintf(buf, sizeof(buf), "pci%d:%d:%d:%d", pci_get_domain(child),
+	    pci_get_bus(child), pci_get_slot(child), pci_get_function(child));
+	return (buf);
+}
+
+/*
+ * Some Host-PCI bridge drivers know which resource ranges they can
+ * decode and should only allocate subranges to child PCI devices.
+ * This API provides a way to manage this.  The bridge drive should
+ * initialize this structure during attach and call
+ * pcib_host_res_decodes() on each resource range it decodes.  It can
+ * then use pcib_host_res_alloc() and pcib_host_res_adjust() as helper
+ * routines for BUS_ALLOC_RESOURCE() and BUS_ADJUST_RESOURCE().  This
+ * API assumes that resources for any decoded ranges can be safely
+ * allocated from the parent via bus_generic_alloc_resource().
+ */
+int
+pcib_host_res_init(device_t pcib, struct pcib_host_resources *hr)
+{
+
+	hr->hr_pcib = pcib;
+	resource_list_init(&hr->hr_rl);
+	return (0);
+}
+
+int
+pcib_host_res_free(device_t pcib, struct pcib_host_resources *hr)
+{
+
+	resource_list_free(&hr->hr_rl);
+	return (0);
+}
+
+int
+pcib_host_res_decodes(struct pcib_host_resources *hr, int type, u_long start,
+    u_long end, u_int flags)
+{
+	struct resource_list_entry *rle;
+	int rid;
+
+	if (bootverbose)
+		device_printf(hr->hr_pcib, "decoding %d %srange %#lx-%#lx\n",
+		    type, flags & RF_PREFETCHABLE ? "prefetchable ": "", start,
+		    end);
+	rid = resource_list_add_next(&hr->hr_rl, type, start, end,
+	    end - start + 1);
+	if (flags & RF_PREFETCHABLE) {
+		KASSERT(type == SYS_RES_MEMORY,
+		    ("only memory is prefetchable"));
+		rle = resource_list_find(&hr->hr_rl, type, rid);
+		rle->flags = RLE_PREFETCH;
+	}
+	return (0);
+}
+
+struct resource *
+pcib_host_res_alloc(struct pcib_host_resources *hr, device_t dev, int type,
+    int *rid, u_long start, u_long end, u_long count, u_int flags)
+{
+	struct resource_list_entry *rle;
+	struct resource *r;
+	u_long new_start, new_end;
+
+	if (flags & RF_PREFETCHABLE)
+		KASSERT(type == SYS_RES_MEMORY,
+		    ("only memory is prefetchable"));
+
+	rle = resource_list_find(&hr->hr_rl, type, 0);
+	if (rle == NULL) {
+		/*
+		 * No decoding ranges for this resource type, just pass
+		 * the request up to the parent.
+		 */
+		return (bus_generic_alloc_resource(hr->hr_pcib, dev, type, rid,
+		    start, end, count, flags));
+	}
+
+restart:
+	/* Try to allocate from each decoded range. */
+	for (; rle != NULL; rle = STAILQ_NEXT(rle, link)) {
+		if (rle->type != type)
+			continue;
+		if (((flags & RF_PREFETCHABLE) != 0) !=
+		    ((rle->flags & RLE_PREFETCH) != 0))
+			continue;
+		new_start = ulmax(start, rle->start);
+		new_end = ulmin(end, rle->end);
+		if (new_start > new_end ||
+		    new_start + count - 1 > new_end ||
+		    new_start + count < new_start)
+			continue;
+		r = bus_generic_alloc_resource(hr->hr_pcib, dev, type, rid,
+		    new_start, new_end, count, flags);
+		if (r != NULL) {
+			if (bootverbose)
+				device_printf(hr->hr_pcib,
+			    "allocated type %d (%#lx-%#lx) for rid %x of %s\n",
+				    type, rman_get_start(r), rman_get_end(r),
+				    *rid, pcib_child_name(dev));
+			return (r);
+		}
+	}
+
+	/*
+	 * If we failed to find a prefetch range for a memory
+	 * resource, try again without prefetch.
+	 */
+	if (flags & RF_PREFETCHABLE) {
+		flags &= ~RF_PREFETCHABLE;
+		rle = resource_list_find(&hr->hr_rl, type, 0);
+		goto restart;
+	}
+	return (NULL);
+}
+
+int
+pcib_host_res_adjust(struct pcib_host_resources *hr, device_t dev, int type,
+    struct resource *r, u_long start, u_long end)
+{
+	struct resource_list_entry *rle;
+
+	rle = resource_list_find(&hr->hr_rl, type, 0);
+	if (rle == NULL) {
+		/*
+		 * No decoding ranges for this resource type, just pass
+		 * the request up to the parent.
+		 */
+		return (bus_generic_adjust_resource(hr->hr_pcib, dev, type, r,
+		    start, end));
+	}
+
+	/* Only allow adjustments that stay within a decoded range. */
+	for (; rle != NULL; rle = STAILQ_NEXT(rle, link)) {
+		if (rle->start <= start && rle->end >= end)
+			return (bus_generic_adjust_resource(hr->hr_pcib, dev,
+			    type, r, start, end));
+	}
+	return (ERANGE);
+}
+#endif /* NEW_PCIB */
diff --git a/sys/dev/pci/pcib_private.h b/sys/dev/pci/pcib_private.h
index 1574deb..0561581 100644
--- a/sys/dev/pci/pcib_private.h
+++ b/sys/dev/pci/pcib_private.h
@@ -33,6 +33,31 @@
 #ifndef __PCIB_PRIVATE_H__
 #define	__PCIB_PRIVATE_H__
 
+#ifdef NEW_PCIB
+/*
+ * Data structure and routines that Host to PCI bridge drivers can use
+ * to restrict allocations for child devices to ranges decoded by the
+ * bridge.
+ */
+struct pcib_host_resources {
+	device_t	hr_pcib;
+	struct resource_list hr_rl;
+};
+
+int		pcib_host_res_init(device_t pcib,
+		    struct pcib_host_resources *hr);
+int		pcib_host_res_free(device_t pcib,
+		    struct pcib_host_resources *hr);
+int		pcib_host_res_decodes(struct pcib_host_resources *hr, int type,
+		    u_long start, u_long end, u_int flags);
+struct resource *pcib_host_res_alloc(struct pcib_host_resources *hr,
+		    device_t dev, int type, int *rid, u_long start, u_long end,
+		    u_long count, u_int flags);
+int		pcib_host_res_adjust(struct pcib_host_resources *hr,
+		    device_t dev, int type, struct resource *r, u_long start,
+		    u_long end);
+#endif
+
 /*
  * Export portions of generic PCI:PCI bridge support so that it can be
  * used by subclasses.
@@ -90,6 +115,9 @@ struct pcib_softc
 
 typedef uint32_t pci_read_config_fn(int b, int s, int f, int reg, int width);
 
+#ifdef NEW_PCIB
+const char	*pcib_child_name(device_t child);
+#endif
 int		host_pcib_get_busno(pci_read_config_fn read_config, int bus,
     int slot, int func, uint8_t *busnum);
 int		pcib_attach(device_t dev);