When numvfs is being updated, reply to numvfs queries with EAGAIN. As drivers may notify changes on numvfs before they are implemented, it is advisable to not reply rather than giving back an outdated value. Clients will notice that the value is still unknown and will retry to get it. The patch relies on the status of the device mutex as an overapproximation of when numvfs is modified. Link: https://bugzilla.kernel.org/show_bug.cgi?id=202991 Signed-off-by: Pierre Crégut <pierre.cregut@xxxxxxxxxx> --- This is an updated version of "PCI/IOV: update num_VFs earlier" and a reply to your mail sent on 14/06/19. Sorry I completely missed it. The commit message was changed as the implementation principle was modified. drivers/pci/pci-sysfs.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c index 965c72104150..493eea261d40 100644 --- a/drivers/pci/pci-sysfs.c +++ b/drivers/pci/pci-sysfs.c @@ -568,6 +568,8 @@ static ssize_t sriov_numvfs_show(struct device *dev, { struct pci_dev *pdev = to_pci_dev(dev); + if (mutex_is_locked(&dev->mutex)) + return -EAGAIN; return sprintf(buf, "%u\n", pdev->sriov->num_VFs); } -- 2.17.1
