Re: [PATCH V1] PCI: designware-ep: Fix DBI access before core init

From: Bjorn Helgaas <helgaas@kernel.org>
To: Vidya Sagar <vidyas@nvidia.com>
Cc: jingoohan1@gmail.com, gustavo.pimentel@synopsys.com,
	lpieralisi@kernel.org, robh@kernel.org, kw@linux.com,
	bhelgaas@google.com, linux-pci@vger.kernel.org,
	linux-kernel@vger.kernel.org, kthota@nvidia.com,
	mmaddireddy@nvidia.com, sagar.tv@gmail.com
Subject: Re: [PATCH V1] PCI: designware-ep: Fix DBI access before core init
Date: Thu, 28 Jul 2022 09:17:16 -0500	[thread overview]
Message-ID: <20220728141716.GA296915@bhelgaas> (raw)
In-Reply-To: <051a3baf-b4dd-7764-2e61-03584cefb4d3@nvidia.com>

On Thu, Jul 28, 2022 at 05:56:28PM +0530, Vidya Sagar wrote:
> 
> 
> On 7/28/2022 3:44 AM, Bjorn Helgaas wrote:
> > External email: Use caution opening links or attachments
> > 
> > 
> > On Wed, Jun 22, 2022 at 09:31:33AM +0530, Vidya Sagar wrote:
> > > Platforms that cannot support their core initialization without the
> > > reference clock from the host, implement the feature 'core_init_notifier'
> > > to indicate the DesignWare sub-system about when their core is getting
> > > initialized. Any accesses to the core (Ex:- DBI) would result in system
> > > hang in such systems (Ex:- tegra194). This patch moves any access to the
> > > core to dw_pcie_ep_init_complete() API which is effectively called only
> > > after the core initialization.
> > 
> > I assume this is still broken.  I want to fix it.  I assume this patch
> > fixes it and there are no known problems with it.  I assume this can
> > be fixed so it works on all platforms, whether they use
> > core_init_notifier or not.
> Yes. All your assumptions are correct.
> 
> > 
> > I'd like the commit log to be specific about where the hang occurs so
> > it's easy for a non-DesignWare expert (me!) to see the problem.  E.g.,
> > on tegra194, X depends on Y, but Y is initialized after X.  Say
> > specifically what functions X and Y are.
> X = DBI accesses
> Y = Core initialization which in turn depends on the REFCLK from the host
> 
> Without this patch, hang happens when DBI registers are accessed without
> core being initialized. In the case of Tegra194 at least, core gets
> initialized only after REFCLK is available from the host. The way we make
> sure that the REFCLK is available from the host is by checking for PERST#
> de-assertion interrupt. (PCIe spec mandates that the host must supply REFCLK
> before de-asserting PERST# signal).
> This patch prevents any accesses to the DBI/Core registers if the platform
> says that it supports core_init_notifier.

That's basically what the commit log already said.  Can you give me
actual function names as a hint for where to look?

Bjorn