Re: [PATCH v3 1/2] PCI: iproc: Retry request when CRS returned from EP

From: Bjorn Helgaas
Date: Mon Jun 19 2017 - 18:39:15 EST

On Tue, Jun 13, 2017 at 09:58:22AM +0530, Oza Oza wrote:
> On Tue, Jun 13, 2017 at 5:00 AM, Bjorn Helgaas <helgaas@xxxxxxxxxx> wrote:
> > Please wrap your changelogs to use 75 columns. "git log" indents the
> > changelog by four spaces, so if your text is 75 wide, it will still
> > fit without wrapping.
> >
> > On Sun, Jun 11, 2017 at 09:35:37AM +0530, Oza Pawandeep wrote:
> >> For Configuration Requests only, following reset
> >> it is possible for a device to terminate the request
> >> but indicate that it is temporarily unable to process
> >> the Request, but will be able to process the Request
> >> in the future â in this case, the Configuration Request
> >> Retry Status 10 (CRS) Completion Status is used
> >
> > How does this relate to the CRS support we already have in the core,
> > e.g., pci_bus_read_dev_vendor_id()? It looks like your root complex
> > already returns 0xffff0001 (CFG_RETRY_STATUS) in some cases.
> >
> > Also, per spec (PCIe r3.1, sec 2.3.2), CRS Software Visibility only
> > affects config reads of the Vendor ID, but you call
> > iproc_pcie_cfg_retry() for all config offsets.
> Yes, as per Spec, CRS Software Visibility only affects config read of
> the Vendor ID.
> For config write or any other config read the Root must automatically
> re-issue configuration
> request again as a new request, and our PCIe RC fails to do so.

OK, if this is a workaround for a hardware defect, let's make that
explicit in the changelog (and probably a comment in the code, too).

I'm actually not sure the spec *requires* the CRS retries to be done
directly in hardware, so it's conceivable the hardware could be
working as designed. But a comment would go a long way toward making
this understandable by differentiating it from the generic CRS
handling in the core.