Re: [PATCH v2 1/2] PCI: dwc: Add host_post_init() callback

[Bjorn Helgaas]

On Tue, Oct 17, 2023 at 01:29:52PM +0530, Manivannan Sadhasivam wrote:
> On Mon, Oct 16, 2023 at 03:58:49PM -0500, Bjorn Helgaas wrote:
> > On Tue, Oct 10, 2023 at 09:29:13PM +0530, Manivannan Sadhasivam wrote:
> > > This callback can be used by the platform drivers to do configuration once
> > > all the devices are scanned. Like changing LNKCTL of all downstream devices
> > > to enable ASPM etc...
> > > 
> > > Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@xxxxxxxxxx>
> > > ---
> > >  drivers/pci/controller/dwc/pcie-designware-host.c | 3 +++
> > >  drivers/pci/controller/dwc/pcie-designware.h      | 1 +
> > >  2 files changed, 4 insertions(+)
> > > 
> > > diff --git a/drivers/pci/controller/dwc/pcie-designware-host.c b/drivers/pci/controller/dwc/pcie-designware-host.c
> > > index a7170fd0e847..7991f0e179b2 100644
> > > --- a/drivers/pci/controller/dwc/pcie-designware-host.c
> > > +++ b/drivers/pci/controller/dwc/pcie-designware-host.c
> > > @@ -502,6 +502,9 @@ int dw_pcie_host_init(struct dw_pcie_rp *pp)
> > >  	if (ret)
> > >  		goto err_stop_link;
> > >  
> > > +	if (pp->ops->host_post_init)
> > > +		pp->ops->host_post_init(pp);
> > 
> > I know we talked about this a little bit in the context of enabling
> > ASPM for devices below qcom 1.9.0 controllers at
> > https://lore.kernel.org/r/20231011050058.GC3508@thinkpad
> > 
> > But I didn't realize at the time that this callback adds a fundamental
> > difference between devices present at boot-time (these devices can be
> > affected by this callback) and any devices hot-added later (we never
> > run this callback again, so anything done by .host_post_init() never
> > applies to them).
> > 
> > We merged this for now, and it helps enable ASPM for builtin devices
> > on qcom, but I don't feel good about this from a larger DWC
> > perspective.  If other drivers use this and they support hot-add, I
> > think we're going to have problems.
> 
> If someone is going to add same ASPM code in host_post_init()
> callback, they will most likely aware of the hotplug issue. I see
> this as an interim solution overall and we should fix the PCI core
> to handle this. But I do not see any straightforward way to enable
> ASPM by default in PCI core as the misbehaving devices can pull the
> system down (atleast in some x86 cases).

Definitely an interim solution, but the interim is going to be long :)

I don't see the PCI core ASPM issue being resolved very soon; it's
been this way ever since the beginning.  I guess there's no point in
me whining without any proposals.

Bjorn