Re: mlx5_ib fails probe with old firmware

[Leon Romanovsky <leon@kernel.org>]

On Tue, Oct 13, 2026 at 10:07:17AM -0400, Tony Battersby wrote:
> On 10/13/26 05:49, Moshe Shemesh wrote:
> >
> > On 10/9/2026 9:20 PM, Tony Battersby wrote:
> >> I have a number of Mellanox infiniband adapters out in the field with
> >> various firmware versions:
> >>
> >> lspci
> >> 01:00.0 Infiniband controller: Mellanox Technologies MT27600 [Connect-IB]
> >> Subsystem: Mellanox Technologies Device 0018
> >>
> >> lspci -n
> >> 01:00.0 0207: 15b3:1011
> >> Subsystem: 15b3:0018
> >>
> >> Ones with old firmware don't work:
> >>
> >> mlx5_core 0000:01:00.0: mlx5_pci_vsc_init:66:(pid 731): Failed to get valid vendor specific ID
> >> mlx5_core 0000:01:00.0: firmware version: 10.10.5020
> >> mlx5_core 0000:01:00.0: 63.008 Gb/s available PCIe bandwidth (8.0 GT/s PCIe x8 link)
> >> mlx5_core 0000:01:00.0: mlx5_cmd_out_err:810:(pid 731): QUERY_ISSI(0x10a) op_mod(0x0) failed, status bad system state(0x4), syndrome (0x408b33), err(-5)
> >> mlx5_core 0000:01:00.0: mlx5_core_set_issi:866:(pid 731): Query ISSI is not supported by FW, ISSI is 0
> >> mlx5_core 0000:01:00.0: mlx5_init_clock:1040:(pid 731): invalid device_frequency_khz, aborting HW clock init
> >> mlx5_core 0000:01:00.0: mlx5_cmd_out_err:810:(pid 731): QUERY_HCA_VPORT_CONTEXT(0x762) op_mod(0x0) failed, status bad operation(0x2), syndrome (0x3590f5), err(-22)
> >> infiniband mlx5_0: Couldn't create per-port data
> >> mlx5_ib.rdma: probe of mlx5_core.rdma.0 failed with error -12
> >>
> >> Ones with new firmware do work:
> >>
> >> mlx5_core 0000:01:00.0: firmware version: 10.12.780
> >> mlx5_core 0000:01:00.0: 63.008 Gb/s available PCIe bandwidth (8.0 GT/s PCIe x8 link)
> >> mlx5_core 0000:01:00.0: mlx5_init_clock:1614:(pid 755): invalid device_frequency_khz, aborting HW clock init
> >> mlx5_core 0000:01:00.0: Flow counters bulk query buffer size increased, bulk_query_len(1)
> >> mlx5_core 0000:01:00.0 mlx5_0: Port: 1 Link INIT
> >> mlx5_core 0000:01:00.0 mlx5_0: Port: 1 Link ACTIVE
> >>
> >> I can work around the problem with the following change:
> >>
> >> diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
> >> index a457647edacd..4fc903c2a0be 100644
> >> --- a/drivers/infiniband/hw/mlx5/main.c
> >> +++ b/drivers/infiniband/hw/mlx5/main.c
> >> @@ -3611,7 +3611,7 @@ static int mlx5_port_immutable(struct ib_device *ibdev, u32 port_num,
> >>   
> >>                  err = mlx5_query_hca_vport_context(dev->mdev, 0, port_num, 0,
> >>                                                     &rep);
> >> -               if (err)
> >> +               if (0 && err)
> >>                          return err;
> >>          }
> >>
> >> This code was added in kernel 4.19 by commit b02289b3d60f ("RDMA:
> >> Validate grh_required when handling AVs").
> >>
> >> A similar problem was fixed in kernel 4.9 by commit f9c14e46748b
> >> ("net/mlx5: Fix query ISSI flow"), which is a source of some of the
> >> messages with the old firmware.
> >>
> >> Can anyone suggest a proper fix for this issue?  I can upgrade the
> >> firmware if necessary, but that would be less convenient than having a
> >> fix for the driver.
> > Hi Tony,
> > Thanks for your report.
> > Please try the following change:
> > diff --git a/drivers/infiniband/hw/mlx5/main.c 
> > b/drivers/infiniband/hw/mlx5/main.c
> > index a457647edacd..632a45f663f6 100644
> > --- a/drivers/infiniband/hw/mlx5/main.c
> > +++ b/drivers/infiniband/hw/mlx5/main.c
> > @@ -3605,7 +3605,8 @@ static int mlx5_port_immutable(struct ib_device 
> > *ibdev, u32 port_num,
> >          if (err)
> >                  return err;
> >
> > -       if (ll == IB_LINK_LAYER_INFINIBAND) {
> > +       if (ll == IB_LINK_LAYER_INFINIBAND &&
> > +           MLX5_CAP_GEN(dev->mdev, ib_virt)) {
> >                  if (ibdev->type == RDMA_DEVICE_TYPE_SMI)
> >                          port_num = smi_to_native_portnum(dev, port_num);
> >
> >
> >
> That does fix the problem with the old firmware.  However,
> the MLX5_CAP_GEN check triggers on both the old and new firmware (I
> added a printk to test). If that is intentional, then it looks good.
> 
> Tested-by: Tony Battersby <tonyb@cybernetics.com>

Thanks, Moshe will submit official patch.

> 
>