Re: [PATCH] PCI: dwc: fix MSI IRQ handler ordering

From: Trent Piepho <tpiepho@impinj.com>
To: "lorenzo.pieralisi@arm.com" <lorenzo.pieralisi@arm.com>,
	"swarren@wwwdotorg.org" <swarren@wwwdotorg.org>,
	"gustavo.pimentel@synopsys.com" <gustavo.pimentel@synopsys.com>
Cc: "jpinto@synopsys.com" <jpinto@synopsys.com>,
	"marex@denx.de" <marex@denx.de>,
	"shawn.guo@linaro.org" <shawn.guo@linaro.org>,
	"jingoohan1@gmail.com" <jingoohan1@gmail.com>,
	"tharvey@gateworks.com" <tharvey@gateworks.com>,
	"m.mann@arkona-technologies.de" <m.mann@arkona-technologies.de>,
	"swarren@nvidia.com" <swarren@nvidia.com>,
	"faiz_abbas@ti.com" <faiz_abbas@ti.com>,
	"xobs@kosagi.com" <xobs@kosagi.com>,
	"hrhaan@gmail.com" <hrhaan@gmail.com>,
	"jbe@pengutronix.de" <jbe@pengutronix.de>,
	"linux-pci@vger.kernel.org" <linux-pci@vger.kernel.org>,
	"bhelgaas@google.com" <bhelgaas@google.com>,
	"arnd@arndb.de" <arnd@arndb.de>
Subject: Re: [PATCH] PCI: dwc: fix MSI IRQ handler ordering
Date: Sat, 5 Jan 2019 00:15:48 +0000	[thread overview]
Message-ID: <1546647347.14247.27.camel@impinj.com> (raw)
In-Reply-To: <253b5dce-08cc-611e-1412-397b9faa5d7a@wwwdotorg.org>

On Fri, 2019-01-04 at 16:23 -0700, Stephen Warren wrote:
> On 1/4/19 4:00 PM, Stephen Warren wrote:
> > 
> > 
> > Hmm. Well I guess I'll go for the patch I posted in our downstream 
> > kernels, since back-porting a bunch of not-yet-available restructuring 
> > to our ancient kernels doesn't sound pleasant:-) But I'll go and take a 
> > quick look at the other patches you mentioned just in case. Thanks!
> 
> So I went and read the thread for "PCI: dwc: Fix interrupt race in when 
> handling MSI" and see:
> 
> Vignesh R wrote:
>  > Lorenzo Pieralisi wrote:
> > > AFAICS:
> > > 
> > > 8c934095fa2f ("PCI: dwc: Clear MSI interrupt status after it is handled,
> > > not before")
> > > 
> > > was fixing a bug, causing "timeouts on some wireless lan cards", we want
> > > to understand what the problem is, fix it once for all on all DWC
> > > based systems.
> > > 
> So it seems quite clear that the correct course of action is to 
> immediately revert commit 8c934095fa2f ("PCI: dwc: Clear MSI interrupt 
> status after it is handled, not before") (or apply my or Trent's patches 
> which are effectively reverts) since it (a) attempts to fix a bug (in 
> the core DWC driver) that doesn't actually exist (the bug is in the DRA 
> HW and is being fixed in the DRA wrapper driver), and (b) has many 
> reports that the change causes regressions; I'm at least the 3rd or 4th 
> person to confirm this now (Faiz originally, then later Trent, Vignesh, 
> myself).

It also seemed clear to me!  But if you check the thread fully, this
was very forcefully rejected.

> Now, whether there's some cleanup or additional fixes needed beyond the 
> simple revert is another question I currently have no insight into, but 
> let's at least get back to a driver without that worked well for people 
> for years, even if there's a theoretical issue to be fixed that nobody 
> hit in practice.

I do not think a real or theoretical mechanism for incorrect behavior
was identified beyond the revert we know of.  More of a case of pieces
not being where the framework intended them to be.