[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAH6h+hczhYdCebrXHnVy4tE6bXGhSJg4GZkfJVYEQtjjb-A-EQ@mail.gmail.com>
Date: Thu, 16 Jan 2020 17:08:31 -0500
From: Marc Smith <msmith626@...il.com>
To: netdev@...r.kernel.org
Subject: 5.4.12 bnxt_en: Unable do read adapter's DSN
Hi,
I'm using the 'bnxt_en' driver with vanilla Linux 5.4.12. I have a
Broadcom P225p 10/25 GbE adapter. I previously used this adapter with
Linux 4.14.120 with no issues. Now with 5.4.12 I observe the following
kernel messages during initialization:
...
[ 2.605878] Broadcom NetXtreme-C/E driver bnxt_en v1.10.0
[ 2.618302] bnxt_en 0000:00:03.0 (unnamed net_device)
(uninitialized): Unable do read adapter's DSN
[ 2.622295] bnxt_en: probe of 0000:00:03.0 failed with error -95
[ 2.632808] bnxt_en 0000:00:0a.0 (unnamed net_device)
(uninitialized): Unable do read adapter's DSN
[ 2.637043] bnxt_en: probe of 0000:00:0a.0 failed with error -95
...
Looks like this comes from bnxt_pcie_dsn_get() in
drivers/net/ethernet/broadcom/bnxt/bnxt.c:
...
int pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_DSN);
u32 dw;
if (!pos) {
netdev_info(bp->dev, "Unable do read adapter's DSN");
return -EOPNOTSUPP;
}
...
And this appears to be related to newer functionality that was not
present in Linux 4.14.x. I am using the P225p via PCI passthrough from
a QEMU/KVM host. I do not have SR-IOV enabled on the card, so not
passing VF's through to the VM (only the PF's).
The card works in the host platform / hypervisor OS (CentOS 7) but
that of course is using the RHEL 3.10.x variant.
I guess my question is:
- Is it expected that a DSN (device serial number) should be available
(fetched via "pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_DSN)) when
the card is passed through into a QEMU/KVM (PCI pass through)? In
other words, is there a problem with my setup/card where the driver
should expect to read a DSN from inside of a virtual machine?
- Or is it normal for a DSN to not be available inside of the virtual
machine guest OS, and the 'bnxt_en' driver needs to be modified to
handle this case? I noticed in the upstream driver (direct from
Broadcom, bnxt_1.10.0-214.0.253.1) that bnxt_pcie_dsn_get() now
resides in bnxt_vfr.c and is called from bnxt_vf_reps_create() (rather
than bnxt_init_one() in 5.4.12).
Any help/advice would be greatly appreciated, thanks in advance.
--Marc
Powered by blists - more mailing lists