From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <linux-pci-owner@vger.kernel.org>
Received: from mail.kernel.org ([198.145.29.136]:41026 "EHLO mail.kernel.org"
	rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
	id S1752566AbbL1W3b (ORCPT <rfc822;linux-pci@vger.kernel.org>);
	Mon, 28 Dec 2015 17:29:31 -0500
Date: Mon, 28 Dec 2015 16:29:28 -0600
From: Bjorn Helgaas <helgaas@kernel.org>
To: Sinan Kaya <okaya@codeaurora.org>
Cc: linux-pci@vger.kernel.org, timur@codeaurora.org,
	cov@codeaurora.org, izumi.taku@jp.fujitsu.com, jcm@redhat.com,
	Bjorn Helgaas <bhelgaas@google.com>,
	Yijing Wang <wangyijing@huawei.com>,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH v2] PCI/AER: enable SERR# forwarding for bridges and
 switches
Message-ID: <20151228222927.GA15808@localhost>
References: <1449074998-9664-1-git-send-email-okaya@codeaurora.org>
 <20151204210622.GI20125@localhost>
 <5669E073.4000801@codeaurora.org>
 <20151210223742.GC367@localhost>
 <566B5CA1.1010306@codeaurora.org>
 <566F08EF.5010904@codeaurora.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
In-Reply-To: <566F08EF.5010904@codeaurora.org>
Sender: linux-pci-owner@vger.kernel.org
List-ID: <linux-pci.vger.kernel.org>

Hi Sinan,

Sorry for the delay in responding; I was on vacation when you sent
this, and I missed it when I returned.

On Mon, Dec 14, 2015 at 01:22:39PM -0500, Sinan Kaya wrote:
> On 12/11/2015 6:30 PM, Sinan Kaya wrote:
> >> I think the best way to fix all the cases would be to do something in
> >> > in pci_configure_device().  Then we could drop the AER bus walk in
> >> > set_downstream_devices_error_reporting().  A bus walk like that is
> >> > always an issue for hotplug.
> >> > 
> > Let me read some code.
> 
> OK, If I understand it right; pci_configure_device is where
> program_hpp_type0 called. You also want to enable AER in this function.
> 
> Move the contents of set_device_error_reporting into
> pci_configure_device like this below ?
> 
> ...
> + int type = pci_pcie_type(dev);
> 
> pci_configure_mps(dev);
> 
> +	if ((type == PCI_EXP_TYPE_ROOT_PORT) ||
> +	    (type == PCI_EXP_TYPE_UPSTREAM) ||
> +	    (type == PCI_EXP_TYPE_DOWNSTREAM)) {
> +		pci_enable_pcie_error_reporting(dev);
> +	}
> 
> +	pcie_set_ecrc_checking(dev);

Yep, that's the sort of thing I'm thinking.

I think there are some subtleties to consider.

_HPP/_HPX can twiddle some of the same bits.  Should we allow _HPP to
clear a bit that pci_enable_pcie_error_reporting() would set?  What
about the reverse?

There are a ridiculous number of places that call
pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR).  We should
do it once and cache the result in struct pci_dev.

Bjorn