Re: [PATCH net-next 3/7] qed: Add support for RoCE hw init

Re: [PATCH net-next 3/7] qed: Add support for RoCE hw init

[Bjorn Helgaas]

[+cc Manish, linux-pci]

On Sat, Oct 01, 2016 at 09:59:57PM +0300, Yuval Mintz wrote:
> From: Ram Amrani <Ram.Amrani@caviumnetworks.com>
> 
> This adds the backbone required for the various HW initalizations
> which are necessary for the qedr driver - FW notification, resource
> initializations, etc.
> ...

> diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
> ...
> +	/* Check atomic operations support in PCI configuration space. */
> +	pci_read_config_dword(cdev->pdev,
> +			      cdev->pdev->pcie_cap + PCI_EXP_DEVCTL2,
> +			      &pci_status_control);
> +
> +	if (pci_status_control & PCI_EXP_DEVCTL2_LTR_EN)
> +		SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_ATOMIC_OP, 1);

I don't understand this.

  1) PCI_EXP_DEVCTL2 is a 16-bit register ("word"), not a 32-bit one
  ("dword").

  2) QED_RDMA_DEV_CAP_ATOMIC_OP is set here but is not read anywhere
  in this patch.  Is it used by the qed device itself?

  3) PCI_EXP_DEVCTL2_LTR_EN is for Latency Tolerance Reporting and is
  not related to atomic ops.  I don't know what
  QED_RDMA_DEV_CAP_ATOMIC_OP means, but possibly one of these was
  intended instead?

    - PCI_EXP_DEVCAP2_ATOMIC_COMP32 means the device supports 32-bit
      AtomicOps as a completer.
    - PCI_EXP_DEVCAP2_ATOMIC_COMP64 means the device supports 64-bit
      AtomicOps as a completer.
    - PCI_EXP_DEVCAP2_ATOMIC_COMP128 means the device supports 128-bit
      AtomicOps as a completer.
    - PCI_EXP_DEVCTL2_ATOMIC_REQ means the device is allowed to
      initiate AtomicOps.

(This code is now in qed_rdma.c)

Re: [PATCH net-next 3/7] qed: Add support for RoCE hw init

[Bjorn Helgaas]

[ping, updated Ariel's address]

On Fri, Oct 07, 2022 at 10:48:32AM -0500, Bjorn Helgaas wrote:
> On Sat, Oct 01, 2016 at 09:59:57PM +0300, Yuval Mintz wrote:
> > From: Ram Amrani <Ram.Amrani@caviumnetworks.com>
> > 
> > This adds the backbone required for the various HW initalizations
> > which are necessary for the qedr driver - FW notification, resource
> > initializations, etc.
> > ...
> 
> > diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c b/drivers/net/ethernet/qlogic/qed/qed_roce.c
> > ...
> > +	/* Check atomic operations support in PCI configuration space. */
> > +	pci_read_config_dword(cdev->pdev,
> > +			      cdev->pdev->pcie_cap + PCI_EXP_DEVCTL2,
> > +			      &pci_status_control);
> > +
> > +	if (pci_status_control & PCI_EXP_DEVCTL2_LTR_EN)
> > +		SET_FIELD(dev->dev_caps, QED_RDMA_DEV_CAP_ATOMIC_OP, 1);
> 
> I don't understand this.
> 
>   1) PCI_EXP_DEVCTL2 is a 16-bit register ("word"), not a 32-bit one
>   ("dword").
> 
>   2) QED_RDMA_DEV_CAP_ATOMIC_OP is set here but is not read anywhere
>   in this patch.  Is it used by the qed device itself?
> 
>   3) PCI_EXP_DEVCTL2_LTR_EN is for Latency Tolerance Reporting and is
>   not related to atomic ops.  I don't know what
>   QED_RDMA_DEV_CAP_ATOMIC_OP means, but possibly one of these was
>   intended instead?
> 
>     - PCI_EXP_DEVCAP2_ATOMIC_COMP32 means the device supports 32-bit
>       AtomicOps as a completer.
>     - PCI_EXP_DEVCAP2_ATOMIC_COMP64 means the device supports 64-bit
>       AtomicOps as a completer.
>     - PCI_EXP_DEVCAP2_ATOMIC_COMP128 means the device supports 128-bit
>       AtomicOps as a completer.
>     - PCI_EXP_DEVCTL2_ATOMIC_REQ means the device is allowed to
>       initiate AtomicOps.
> 
> (This code is now in qed_rdma.c)

RE: [PATCH net-next 3/7] qed: Add support for RoCE hw init

[Michal Kalderon]

> From: Bjorn Helgaas <helgaas@kernel.org>
> Sent: Tuesday, October 11, 2022 12:45 AM
> 
> [ping, updated Ariel's address]
> 
> On Fri, Oct 07, 2022 at 10:48:32AM -0500, Bjorn Helgaas wrote:
> > On Sat, Oct 01, 2016 at 09:59:57PM +0300, Yuval Mintz wrote:
> > > From: Ram Amrani <Ram.Amrani@caviumnetworks.com>
> > >
> > > This adds the backbone required for the various HW initalizations
> > > which are necessary for the qedr driver - FW notification, resource
> > > initializations, etc.
> > > ...
> >
> > > diff --git a/drivers/net/ethernet/qlogic/qed/qed_roce.c
> b/drivers/net/ethernet/qlogic/qed/qed_roce.c
> > > ...
> > > +	/* Check atomic operations support in PCI configuration space. */
> > > +	pci_read_config_dword(cdev->pdev,
> > > +			      cdev->pdev->pcie_cap + PCI_EXP_DEVCTL2,
> > > +			      &pci_status_control);
> > > +
> > > +	if (pci_status_control & PCI_EXP_DEVCTL2_LTR_EN)
> > > +		SET_FIELD(dev->dev_caps,
> QED_RDMA_DEV_CAP_ATOMIC_OP, 1);
> >
> > I don't understand this.
> >
> >   1) PCI_EXP_DEVCTL2 is a 16-bit register ("word"), not a 32-bit one
> >   ("dword").
> >
> >   2) QED_RDMA_DEV_CAP_ATOMIC_OP is set here but is not read
> anywhere
> >   in this patch.  Is it used by the qed device itself?
> >
> >   3) PCI_EXP_DEVCTL2_LTR_EN is for Latency Tolerance Reporting and is
> >   not related to atomic ops.  I don't know what
> >   QED_RDMA_DEV_CAP_ATOMIC_OP means, but possibly one of these
> was
> >   intended instead?
> >
> >     - PCI_EXP_DEVCAP2_ATOMIC_COMP32 means the device supports 32-
> bit
> >       AtomicOps as a completer.
> >     - PCI_EXP_DEVCAP2_ATOMIC_COMP64 means the device supports 64-
> bit
> >       AtomicOps as a completer.
> >     - PCI_EXP_DEVCAP2_ATOMIC_COMP128 means the device supports
> 128-bit
> >       AtomicOps as a completer.
> >     - PCI_EXP_DEVCTL2_ATOMIC_REQ means the device is allowed to
> >       initiate AtomicOps.
> >
> > (This code is now in qed_rdma.c)

Thanks for looking into this. 
This seems like redundant code and left-overs for supporting the atomic operation verb.
Atomic support is handled by: qedr_pci_set_atomic and introduced with a proper implementation by commit-SHA 20c3ff6114b0c

This code in qed_rdma.c can safely be removed.
Thanks,
Michal