[dpdk-dev] [PATCH v2 3/3] bus/pci: only consider usable devices to select IOVA mode

[David Marchand <david.marchand@redhat.com>]

From: Ben Walker <benjamin.walker@intel.com>

When selecting the preferred IOVA mode of the pci bus, the current
heuristic ("are devices bound?", "are devices bound to UIO?", "are pmd
drivers supporting IOVA as VA?" etc..) should honor the device
white/blacklist so that an unwanted device does not impact the decision.

There is no reason to consider a device which has no driver available.

This applies to all OS, so implements this in common code then call a
OS specific callback.

On Linux side:
- the VFIO special considerations should be evaluated only if VFIO
  support is built,
- there is no strong requirement on using VA rather than PA if a driver
  supports VA, so defaulting to DC in such a case.

Signed-off-by: Ben Walker <benjamin.walker@intel.com>
Signed-off-by: David Marchand <david.marchand@redhat.com>
---
 drivers/bus/pci/bsd/pci.c    |   9 +-
 drivers/bus/pci/linux/pci.c  | 191 ++++++++++++-------------------------------
 drivers/bus/pci/pci_common.c |  65 +++++++++++++++
 drivers/bus/pci/private.h    |   8 ++
 4 files changed, 131 insertions(+), 142 deletions(-)

diff --git a/drivers/bus/pci/bsd/pci.c b/drivers/bus/pci/bsd/pci.c
index c7b90cb..a2de709 100644
--- a/drivers/bus/pci/bsd/pci.c
+++ b/drivers/bus/pci/bsd/pci.c
@@ -376,13 +376,14 @@
 	return -1;
 }
 
-/*
- * Get iommu class of PCI devices on the bus.
- */
 enum rte_iova_mode
-rte_pci_get_iommu_class(void)
+pci_device_iova_mode(const struct rte_pci_driver *pdrv __rte_unused,
+		     const struct rte_pci_device *pdev)
 {
 	/* Supports only RTE_KDRV_NIC_UIO */
+	if (pdev->kdrv != RTE_KDRV_NIC_UIO)
+		RTE_LOG(DEBUG, EAL, "Unsupported kernel driver? Defaulting to IOVA as 'PA'\n");
+
 	return RTE_IOVA_PA;
 }
 
diff --git a/drivers/bus/pci/linux/pci.c b/drivers/bus/pci/linux/pci.c
index b931cf9..33c8ea7 100644
--- a/drivers/bus/pci/linux/pci.c
+++ b/drivers/bus/pci/linux/pci.c
@@ -500,95 +500,14 @@
 	return -1;
 }
 
-/*
- * Is pci device bound to any kdrv
- */
-static inline int
-pci_one_device_is_bound(void)
-{
-	struct rte_pci_device *dev = NULL;
-	int ret = 0;
-
-	FOREACH_DEVICE_ON_PCIBUS(dev) {
-		if (dev->kdrv == RTE_KDRV_UNKNOWN ||
-		    dev->kdrv == RTE_KDRV_NONE) {
-			continue;
-		} else {
-			ret = 1;
-			break;
-		}
-	}
-	return ret;
-}
-
-/*
- * Any one of the device bound to uio
- */
-static inline int
-pci_one_device_bound_uio(void)
-{
-	struct rte_pci_device *dev = NULL;
-	struct rte_devargs *devargs;
-	int need_check;
-
-	FOREACH_DEVICE_ON_PCIBUS(dev) {
-		devargs = dev->device.devargs;
-
-		need_check = 0;
-		switch (rte_pci_bus.bus.conf.scan_mode) {
-		case RTE_BUS_SCAN_WHITELIST:
-			if (devargs && devargs->policy == RTE_DEV_WHITELISTED)
-				need_check = 1;
-			break;
-		case RTE_BUS_SCAN_UNDEFINED:
-		case RTE_BUS_SCAN_BLACKLIST:
-			if (devargs == NULL ||
-			    devargs->policy != RTE_DEV_BLACKLISTED)
-				need_check = 1;
-			break;
-		}
-
-		if (!need_check)
-			continue;
-
-		if (dev->kdrv == RTE_KDRV_IGB_UIO ||
-		   dev->kdrv == RTE_KDRV_UIO_GENERIC) {
-			return 1;
-		}
-	}
-	return 0;
-}
-
-/*
- * Any one of the device has iova as va
- */
-static inline int
-pci_one_device_has_iova_va(void)
-{
-	struct rte_pci_device *dev = NULL;
-	struct rte_pci_driver *drv = NULL;
-
-	FOREACH_DRIVER_ON_PCIBUS(drv) {
-		if (drv && drv->drv_flags & RTE_PCI_DRV_IOVA_AS_VA) {
-			FOREACH_DEVICE_ON_PCIBUS(dev) {
-				if ((dev->kdrv == RTE_KDRV_VFIO ||
-				     dev->kdrv == RTE_KDRV_NIC_MLX) &&
-				    rte_pci_match(drv, dev))
-					return 1;
-			}
-		}
-	}
-	return 0;
-}
-
 #if defined(RTE_ARCH_X86)
 static bool
-pci_one_device_iommu_support_va(struct rte_pci_device *dev)
+pci_one_device_iommu_support_va(const struct rte_pci_device *dev)
 {
 #define VTD_CAP_MGAW_SHIFT	16
 #define VTD_CAP_MGAW_MASK	(0x3fULL << VTD_CAP_MGAW_SHIFT)
 #define X86_VA_WIDTH 47 /* From Documentation/x86/x86_64/mm.txt */
-	struct rte_pci_addr *addr = &dev->addr;
+	const struct rte_pci_addr *addr = &dev->addr;
 	char filename[PATH_MAX];
 	FILE *fp;
 	uint64_t mgaw, vtd_cap_reg = 0;
@@ -632,80 +551,76 @@
 }
 #elif defined(RTE_ARCH_PPC_64)
 static bool
-pci_one_device_iommu_support_va(__rte_unused struct rte_pci_device *dev)
+pci_one_device_iommu_support_va(__rte_unused const struct rte_pci_device *dev)
 {
 	return false;
 }
 #else
 static bool
-pci_one_device_iommu_support_va(__rte_unused struct rte_pci_device *dev)
+pci_one_device_iommu_support_va(__rte_unused const struct rte_pci_device *dev)
 {
 	return true;
 }
 #endif
 
-/*
- * All devices IOMMUs support VA as IOVA
- */
-static bool
-pci_devices_iommu_support_va(void)
+enum rte_iova_mode
+pci_device_iova_mode(const struct rte_pci_driver *pdrv,
+		     const struct rte_pci_device *pdev)
 {
-	struct rte_pci_device *dev = NULL;
-	struct rte_pci_driver *drv = NULL;
+	enum rte_iova_mode iova_mode = RTE_IOVA_DC;
+	static int iommu_no_va = -1;
 
-	FOREACH_DRIVER_ON_PCIBUS(drv) {
-		FOREACH_DEVICE_ON_PCIBUS(dev) {
-			if (!rte_pci_match(drv, dev))
-				continue;
-			/*
-			 * just one PCI device needs to be checked out because
-			 * the IOMMU hardware is the same for all of them.
-			 */
-			return pci_one_device_iommu_support_va(dev);
+	switch (pdev->kdrv) {
+	case RTE_KDRV_VFIO: {
+#ifdef VFIO_PRESENT
+		static int is_vfio_noiommu_enabled = -1;
+
+		if (is_vfio_noiommu_enabled == -1) {
+			if (rte_vfio_noiommu_is_enabled() == 1)
+				is_vfio_noiommu_enabled = 1;
+			else
+				is_vfio_noiommu_enabled = 0;
+		}
+		if ((pdrv->drv_flags & RTE_PCI_DRV_IOVA_AS_VA) == 0) {
+			iova_mode = RTE_IOVA_PA;
+		} else if (is_vfio_noiommu_enabled != 0) {
+			RTE_LOG(DEBUG, EAL, "Forcing to 'PA', vfio-noiommu mode configured\n");
+			iova_mode = RTE_IOVA_PA;
 		}
+#endif
+		break;
 	}
-	return true;
-}
 
-/*
- * Get iommu class of PCI devices on the bus.
- */
-enum rte_iova_mode
-rte_pci_get_iommu_class(void)
-{
-	bool is_bound;
-	bool is_vfio_noiommu_enabled = true;
-	bool has_iova_va;
-	bool is_bound_uio;
-	bool iommu_no_va;
-
-	is_bound = pci_one_device_is_bound();
-	if (!is_bound)
-		return RTE_IOVA_DC;
-
-	has_iova_va = pci_one_device_has_iova_va();
-	is_bound_uio = pci_one_device_bound_uio();
-	iommu_no_va = !pci_devices_iommu_support_va();
-#ifdef VFIO_PRESENT
-	is_vfio_noiommu_enabled = rte_vfio_noiommu_is_enabled() == true ?
-					true : false;
-#endif
+	case RTE_KDRV_NIC_MLX:
+		if ((pdrv->drv_flags & RTE_PCI_DRV_IOVA_AS_VA) == 0)
+			iova_mode = RTE_IOVA_PA;
+		break;
 
-	if (has_iova_va && !is_bound_uio && !is_vfio_noiommu_enabled &&
-			!iommu_no_va)
-		return RTE_IOVA_VA;
+	case RTE_KDRV_IGB_UIO:
+	case RTE_KDRV_UIO_GENERIC:
+		iova_mode = RTE_IOVA_PA;
+		break;
 
-	if (has_iova_va) {
-		RTE_LOG(WARNING, EAL, "Some devices want iova as va but pa will be used because.. ");
-		if (is_vfio_noiommu_enabled)
-			RTE_LOG(WARNING, EAL, "vfio-noiommu mode configured\n");
-		if (is_bound_uio)
-			RTE_LOG(WARNING, EAL, "few device bound to UIO\n");
-		if (iommu_no_va)
-			RTE_LOG(WARNING, EAL, "IOMMU does not support IOVA as VA\n");
+	default:
+		RTE_LOG(DEBUG, EAL, "Unsupported kernel driver? Defaulting to IOVA as 'PA'\n");
+		iova_mode = RTE_IOVA_PA;
+		break;
 	}
 
-	return RTE_IOVA_PA;
+	if (iova_mode != RTE_IOVA_PA) {
+		/*
+		 * We can check this only once, because the IOMMU hardware is
+		 * the same for all of them.
+		 */
+		if (iommu_no_va == -1)
+			iommu_no_va = pci_one_device_iommu_support_va(pdev)
+					? 0 : 1;
+		if (iommu_no_va != 0) {
+			RTE_LOG(DEBUG, EAL, "Forcing to 'PA', IOMMU does not support IOVA as 'VA'\n");
+			iova_mode = RTE_IOVA_PA;
+		}
+	}
+	return iova_mode;
 }
 
 /* Read PCI config space. */
diff --git a/drivers/bus/pci/pci_common.c b/drivers/bus/pci/pci_common.c
index 704b9d7..d2af472 100644
--- a/drivers/bus/pci/pci_common.c
+++ b/drivers/bus/pci/pci_common.c
@@ -574,6 +574,71 @@ static struct rte_devargs *pci_devargs_lookup(struct rte_pci_device *dev)
 	return -1;
 }
 
+static bool
+pci_ignore_device(const struct rte_pci_device *dev)
+{
+	struct rte_devargs *devargs = dev->device.devargs;
+
+	switch (rte_pci_bus.bus.conf.scan_mode) {
+	case RTE_BUS_SCAN_WHITELIST:
+		if (devargs && devargs->policy == RTE_DEV_WHITELISTED)
+			return false;
+		break;
+	case RTE_BUS_SCAN_UNDEFINED:
+	case RTE_BUS_SCAN_BLACKLIST:
+		if (devargs == NULL ||
+		    devargs->policy != RTE_DEV_BLACKLISTED)
+			return false;
+		break;
+	}
+	return true;
+}
+
+enum rte_iova_mode
+rte_pci_get_iommu_class(void)
+{
+	enum rte_iova_mode iova_mode = RTE_IOVA_DC;
+	const struct rte_pci_device *dev;
+	const struct rte_pci_driver *drv;
+	bool devices_want_va = false;
+	bool devices_want_pa = false;
+
+	FOREACH_DEVICE_ON_PCIBUS(dev) {
+		if (pci_ignore_device(dev))
+			continue;
+		if (dev->kdrv == RTE_KDRV_UNKNOWN ||
+		    dev->kdrv == RTE_KDRV_NONE)
+			continue;
+		FOREACH_DRIVER_ON_PCIBUS(drv) {
+			enum rte_iova_mode dev_iova_mode;
+
+			if (!rte_pci_match(drv, dev))
+				continue;
+
+			dev_iova_mode = pci_device_iova_mode(drv, dev);
+			RTE_LOG(DEBUG, EAL, "PCI driver %s for device "
+				PCI_PRI_FMT " wants IOVA as '%s'\n",
+				drv->driver.name,
+				dev->addr.domain, dev->addr.bus,
+				dev->addr.devid, dev->addr.function,
+				dev_iova_mode == RTE_IOVA_DC ? "DC" :
+				(dev_iova_mode == RTE_IOVA_PA ? "PA" : "VA"));
+			if (dev_iova_mode == RTE_IOVA_PA)
+				devices_want_pa = true;
+			else if (dev_iova_mode == RTE_IOVA_VA)
+				devices_want_va = true;
+		}
+	}
+	if (devices_want_pa) {
+		iova_mode = RTE_IOVA_PA;
+		if (devices_want_va)
+			RTE_LOG(WARNING, EAL, "Some devices want 'VA' but forcing 'PA' because other devices want it\n");
+	} else if (devices_want_va) {
+		iova_mode = RTE_IOVA_VA;
+	}
+	return iova_mode;
+}
+
 struct rte_pci_bus rte_pci_bus = {
 	.bus = {
 		.scan = rte_pci_scan,
diff --git a/drivers/bus/pci/private.h b/drivers/bus/pci/private.h
index 13c3324..8a55240 100644
--- a/drivers/bus/pci/private.h
+++ b/drivers/bus/pci/private.h
@@ -173,6 +173,14 @@ int pci_uio_map_resource_by_index(struct rte_pci_device *dev, int res_idx,
 	      const struct rte_pci_device *pci_dev);
 
 /**
+ * OS specific callback for rte_pci_get_iommu_class
+ *
+ */
+enum rte_iova_mode
+pci_device_iova_mode(const struct rte_pci_driver *pci_drv,
+		     const struct rte_pci_device *pci_dev);
+
+/**
  * Get iommu class of PCI devices on the bus.
  * And return their preferred iova mapping mode.
  *
-- 
1.8.3.1