mlx4 fails ("go bit not cleared") without MSI

mlx4 fails ("go bit not cleared") without MSI

[Bjorn Helgaas]

Hi Eugenia,

Due to my own stupidity, I tried to use mlx4 without turning on
CONFIG_PCI_MSI=y.  The driver failed miserably:

  pci 0000:01:00.0: [15b3:1007] type 00 class 0x020000
  mlx4_core: Mellanox ConnectX core driver v2.2-1 (Feb, 2014)
  mlx4_core: Initializing 0000:01:00.0
  mlx4_core 0000:01:00.0: PCIe link speed is 8.0GT/s, device supports 8.0GT/s
  mlx4_core 0000:01:00.0: PCIe link width is x8, device supports x8
  mlx4_core 0000:01:00.0: command 0x23 timed out (go bit not cleared)
  mlx4_core 0000:01:00.0: device is going to be reset
  mlx4_core 0000:01:00.0: device was reset successfully
  mlx4_core 0000:01:00.0: Failed to initialize queue pair table, aborting
  mlx4_core: probe of 0000:01:00.0 failed with error -5

The lack of MSI was my own fault, but it took me a long time to figure it
out.

The fact that mlx4 has a "msi_x" module parameter makes me think that
mlx4 is *supposed* to work without MSI.  Would you consider either
fixing the non-MSI fallback path, or printing a warning when the
driver can't set up MSIs?

This is easy to reproduce; just turn off CONFIG_PCI_MSI or boot with
"pci=nomsi".

Bjorn

RE: mlx4 fails ("go bit not cleared") without MSI

[Yishai Hadas]

> From: Bjorn Helgaas [mailto:bhelgaas@google.com]
> I assume you're tried booting with "pci=nomsi" or "mlx_core.msi_x=0"
> and it works for you?  Note that my board might be an internal design,
> not a Mellanox board, so if it works for you, we can probably just
> ignore this as being an artifact of the internal design.

Yes, it works for me (module parameter msi_x=0), as you pointed out above
you are not using Mellanox board but some internal design and
it might be some artifact of.

Re: mlx4 fails ("go bit not cleared") without MSI

[Bjorn Helgaas]

On Wed, Jan 20, 2016 at 3:42 AM, Yishai Hadas <yishaih@mellanox.com> wrote:
>> From: Bjorn Helgaas [mailto:bhelgaas@google.com]
>> I assume you're tried booting with "pci=nomsi" or "mlx_core.msi_x=0"
>> and it works for you?  Note that my board might be an internal design,
>> not a Mellanox board, so if it works for you, we can probably just
>> ignore this as being an artifact of the internal design.
>
> Yes, it works for me (module parameter msi_x=0), as you pointed out above
> you are not using Mellanox board but some internal design and
> it might be some artifact of.

OK, let's assume that's the case.  I haven't seen the actual board, so
I can't be certain what I'm using.  Thanks for checking into it.

Bjorn