Re: [PATCH v2 1/7] PCI/ASPM: Cache device's ASPM link capability in struct pci_dev

From: Bjorn Helgaas <helgaas@kernel.org>
To: "Saheed O. Bolarinwa" <refactormyself@gmail.com>
Cc: linux-pci@vger.kernel.org
Subject: Re: [PATCH v2 1/7] PCI/ASPM: Cache device's ASPM link capability in struct pci_dev
Date: Thu, 24 Sep 2020 17:53:47 -0500	[thread overview]
Message-ID: <20200924225347.GA2367309@bjorn-Precision-5520> (raw)
In-Reply-To: <20200924142443.260861-2-refactormyself@gmail.com>

On Thu, Sep 24, 2020 at 04:24:37PM +0200, Saheed O. Bolarinwa wrote:
> pcie_get_aspm_reg() reads LNKCAP to learn whether the device supports
> ASPM L0s and/or L1 and L1 substates.
> 
> If we cache the entire LNKCAP word early enough, we may be able to
> use it in other places that read LNKCAP, e.g. pcie_get_speed_cap(),
> pcie_get_width_cap(), pcie_init(), etc.
> 
>  - Add struct pci_dev.lnkcap (u32)
>  - Read PCI_EXP_LNKCAP in set_pcie_port_type() and save it
>    in pci_dev.lnkcap
>  - Use pdev->lnkcap instead of reading PCI_EXP_LNKCAP

I think we need to be a little careful here because there's a note in
the spec (PCIe r5.0, sec 7.5.3.6):

  Note that exit latencies may be influenced by PCI Express reference
  clock configuration depending upon whether a component uses a common
  or separate reference clock.

So if we change the common clock configuration, e.g., in
pcie_aspm_configure_common_clock() or anything else that writes
PCI_EXP_LNKCTL_CCC, I think we will need to update pdev->lnkcap.