Re: [PATCH] PCI/IOV: update num_VFs earlier

From: Don Dutile
Date: Wed Oct 09 2019 - 10:20:37 EST


On 10/09/2019 08:31 AM, Bjorn Helgaas wrote:
On Tue, Oct 08, 2019 at 06:06:46PM -0400, Don Dutile wrote:
On 10/08/2019 05:38 PM, Bjorn Helgaas wrote:
On Thu, Oct 03, 2019 at 05:10:07PM -0500, Bjorn Helgaas wrote:
On Thu, Oct 03, 2019 at 11:04:45AM +0200, CREGUT Pierre IMT/OLN wrote:
...

NIC drivers send netlink events when their state change, but it is
the core that changes the value of num_vfs. So I would think it is
the core responsibility to make sure the exposed value makes sense
and it would be better to ignore the details of the driver
implementation.

Yes, I think you're right. And I like your previous suggestion of
just locking the device in the reader. I'm not enough of a sysfs
expert to know if there's a good reason to avoid a lock there. Does
the following look reasonable to you?

I applied the patch below to pci/virtualization for v5.5, thanks for
I hope not... see below

your great patience!

commit 0940fc95da45
Author: Pierre CrÃgut <pierre.cregut@xxxxxxxxxx>
Date: Wed Sep 11 09:27:36 2019 +0200

PCI/IOV: Serialize sysfs sriov_numvfs reads vs writes
When sriov_numvfs is being updated, drivers may notify about new devices
before they are reflected in sriov->num_VFs, so concurrent sysfs reads
previously returned stale values.
Serialize the sysfs read vs the write so the read returns the correct
num_VFs value.
Link: https://bugzilla.kernel.org/show_bug.cgi?id=202991
Link: https://lore.kernel.org/r/20190911072736.32091-1-pierre.cregut@xxxxxxxxxx
Signed-off-by: Pierre CrÃgut <pierre.cregut@xxxxxxxxxx>
Signed-off-by: Bjorn Helgaas <bhelgaas@xxxxxxxxxx>

diff --git a/drivers/pci/iov.c b/drivers/pci/iov.c
index b3f972e8cfed..e77562aabbae 100644
--- a/drivers/pci/iov.c
+++ b/drivers/pci/iov.c
@@ -254,8 +254,14 @@ static ssize_t sriov_numvfs_show(struct device *dev,
char *buf)
{
struct pci_dev *pdev = to_pci_dev(dev);
+ u16 num_vfs;
+
+ /* Serialize vs sriov_numvfs_store() so readers see valid num_VFs */
+ device_lock(&pdev->dev);
^^^^^ lock
+ num_vfs = pdev->sriov->num_VFs;
+ device_lock(&pdev->dev);
^^^^ and lock again!

Oops, sorry, my fault. Fixed.

Thanks.
--dd

- return sprintf(buf, "%u\n", pdev->sriov->num_VFs);
+ return sprintf(buf, "%u\n", num_vfs);
}
/*