Received-SPF: pass (google.com: domain of linux-kernel+bounces-216020-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
Date: Sat, 15 Jun 2024 16:26:03 -0500
From: Bjorn Helgaas <helgaas@kernel.org>
To: Songyang Li <leesongyang@outlook.com>
Cc: bhelgaas@google.com, linux-kernel@vger.kernel.org,
	linux-pci@vger.kernel.org
Subject: Re: [PATCH] PCI: Cancel compilation restrictions on function
 pcie_clear_device_status
Message-ID: <20240615212603.GA1157372@bhelgaas>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <TY3P286MB2754862E7F9F4DE1B25BF357B4C32@TY3P286MB2754.JPNP286.PROD.OUTLOOK.COM>

On Sat, Jun 15, 2024 at 11:13:07AM +0800, Songyang Li wrote:
> On Wed, 12 Jun 2024 15:14:32 -0500, Bjorn Helgaas wrote:
> > I think all current any callers of pcie_clear_device_status() are also
> > under CONFIG_PCIEAER, so I don't think this fixes a current problem.
> > 
> > As you point out, it might make sense to use
> > pcie_clear_device_status() even without AER, but I think we should
> > include this change at the time when we add such a use.
> > 
> > If I'm missing a use with the current kernel, let me know.
> 
> As far as I know, some PCIe device drivers, for example,
> [net/ethernet/broadcom/tg3.c],[net/ethernet/atheros/atl1c/atl1c_main.c],
> which use the following code to clear the device status register,
> pcie_capability_write_word(tp->pdev, PCI_EXP_DEVSTA,
>                 PCI_EXP_DEVSTA_CED |
>                 PCI_EXP_DEVSTA_NFED |
>                 PCI_EXP_DEVSTA_FED |
>                 PCI_EXP_DEVSTA_URD);
> I think it may be more suitable to export the pcie_clear_device_status()
> for use in the driver code.

If we want to use this from drivers, it would make sense to do
something like this patch, and this patch could be part of a series to
call it from the drivers.

But at the same time, we should ask whether drivers should be clearing
this status themselves, or whether it should be done by the PCI core.

Bjorn