git: 4b195f1a25d5 - main - pci_iov: Roll back failed VF enumeration

The branch main has been updated by kbowling:

URL: https://cgit.FreeBSD.org/src/commit/?id=4b195f1a25d5003117653a1a323ad19561dc8705

commit 4b195f1a25d5003117653a1a323ad19561dc8705
Author:     Kevin Bowling <[email protected]>
AuthorDate: 2026-08-10 04:53:55 +0000
Commit:     Kevin Bowling <[email protected]>
CommitDate: 2026-08-10 08:47:59 +0000

    pci_iov: Roll back failed VF enumeration
    
    pci_iov_enumerate_vfs() logged a failed VF creation or driver
    configuration but still reported the whole SR-IOV configuration as
    successful.  The PF remained enabled with the requested NumVFs and
    driver state even though one or more VF children were absent.
    
    Make VF enumeration atomic.  Delete children created by the failed
    attempt, invoke the PF driver cleanup, disable VF memory space and VF
    Enable, release the IOV resources, and return the original error to
    iovctl.  Also treat failure to create a VF child as an error instead
    of silently accepting a partial configuration.
    
    MFC after:      2 weeks
---
 sys/dev/pci/pci_iov.c | 33 ++++++++++++++++++++++++++-------
 1 file changed, 26 insertions(+), 7 deletions(-)

diff --git a/sys/dev/pci/pci_iov.c b/sys/dev/pci/pci_iov.c
index 545d91e372f4..5d39de605c22 100644
--- a/sys/dev/pci/pci_iov.c
+++ b/sys/dev/pci/pci_iov.c
@@ -619,16 +619,16 @@ pci_iov_setup_bars(struct pci_devinfo *dinfo)
 	return (0);
 }
 
-static void
+static int
 pci_iov_enumerate_vfs(struct pci_devinfo *dinfo, const nvlist_t *config,
     uint16_t first_rid, uint16_t rid_stride)
 {
 	char device_name[VF_MAX_NAME];
 	const nvlist_t *device, *driver_config, *iov_config;
-	device_t bus, dev, vf;
+	device_t bus, dev, vf, *vfs;
 	struct pcicfg_iov *iov;
 	struct pci_devinfo *vfinfo;
-	int i, error;
+	int error, i, ncreated;
 	uint16_t vid, did, next_rid;
 
 	iov = dinfo->cfg.iov;
@@ -637,6 +637,9 @@ pci_iov_enumerate_vfs(struct pci_devinfo *dinfo, const nvlist_t *config,
 	next_rid = first_rid;
 	vid = pci_get_vendor(dev);
 	did = IOV_READ(dinfo, PCIR_SRIOV_VF_DID, 2);
+	vfs = mallocarray(iov->iov_num_vfs, sizeof(*vfs), M_SRIOV,
+	    M_WAITOK | M_ZERO);
+	ncreated = 0;
 
 	for (i = 0; i < iov->iov_num_vfs; i++, next_rid += rid_stride) {
 		snprintf(device_name, sizeof(device_name), VF_PREFIX"%d", i);
@@ -645,8 +648,11 @@ pci_iov_enumerate_vfs(struct pci_devinfo *dinfo, const nvlist_t *config,
 		driver_config = nvlist_get_nvlist(device, DRIVER_CONFIG_NAME);
 
 		vf = PCI_CREATE_IOV_CHILD(bus, dev, next_rid, vid, did);
-		if (vf == NULL)
-			break;
+		if (vf == NULL) {
+			error = ENXIO;
+			goto fail;
+		}
+		vfs[ncreated++] = vf;
 
 		/*
 		 * If we are creating passthrough devices then force the ppt
@@ -666,11 +672,19 @@ pci_iov_enumerate_vfs(struct pci_devinfo *dinfo, const nvlist_t *config,
 		error = PCI_IOV_ADD_VF(dev, i, driver_config);
 		if (error != 0) {
 			device_printf(dev, "Failed to add VF %d\n", i);
-			device_delete_child(bus, vf);
+			goto fail;
 		}
 	}
 
 	bus_attach_children(bus);
+	free(vfs, M_SRIOV);
+	return (0);
+
+fail:
+	for (i = 0; i < ncreated; i++)
+		device_delete_child(bus, vfs[i]);
+	free(vfs, M_SRIOV);
+	return (error);
 }
 
 static int
@@ -799,7 +813,9 @@ pci_iov_config(struct cdev *cdev, struct pci_iov_arg *arg)
 
 	/* Per specification, we must wait 100ms before accessing VFs. */
 	pause("iov", roundup(hz, 10));
-	pci_iov_enumerate_vfs(dinfo, config, first_rid, rid_stride);
+	error = pci_iov_enumerate_vfs(dinfo, config, first_rid, rid_stride);
+	if (error != 0)
+		goto out;
 
 	nvlist_destroy(config);
 	iov->iov_flags &= ~IOV_BUSY;
@@ -809,6 +825,9 @@ pci_iov_config(struct cdev *cdev, struct pci_iov_arg *arg)
 out:
 	if (iov_inited) {
 		PCI_IOV_UNINIT(dev);
+		iov_ctl = IOV_READ(dinfo, PCIR_SRIOV_CTL, 2);
+		iov_ctl &= ~(PCIM_SRIOV_VF_EN | PCIM_SRIOV_VF_MSE);
+		IOV_WRITE(dinfo, PCIR_SRIOV_CTL, iov_ctl, 2);
 		IOV_WRITE(dinfo, PCIR_SRIOV_NUM_VFS, 0, 2);
 	}