Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Mon, 1 Jun 2020 10:13:39 -0500
From:   Bjorn Helgaas <helgaas@kernel.org>
To:     Christoph Hellwig <hch@infradead.org>
Cc:     Matthew Garrett <mjg59@srcf.ucam.org>,
        Heiner Kallweit <hkallweit1@gmail.com>,
        Bjorn Helgaas <bhelgaas@google.com>,
        "Rafael J. Wysocki" <rjw@rjwysocki.net>,
        "linux-pci@vger.kernel.org" <linux-pci@vger.kernel.org>,
        linux-kernel@vger.kernel.org, linux-acpi@vger.kernel.org
Subject: Re: Lost PCIe PME after a914ff2d78ce ("PCI/ASPM: Don't select
 CONFIG_PCIEASPM by default")
Message-ID: <20200530123421.GA532471@bjorn-Precision-5520>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200530113344.GA2834@infradead.org>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sat, May 30, 2020 at 04:33:44AM -0700, Christoph Hellwig wrote:
> On Sat, May 30, 2020 at 08:14:34AM +0100, Matthew Garrett wrote:
> > On Sat, May 30, 2020 at 08:33:50AM +0200, Heiner Kallweit wrote:
> > 
> > > It *was* default y. This changed with a914ff2d78ce ("PCI/ASPM: Don't
> > > select CONFIG_PCIEASPM by default") and that's what triggered the
> > > problem. If there's no easy solution, then maybe it's best to revert
> > > the change for now.
> > 
> > Oh, sorry, I was looking at mainline. CONFIG_PCIEASPM should 
> > *definitely* be enabled by default - platforms expect the OS to support 
> > it. If we want to get rid of default y then I think it'd make more sense 
> > to have a CONFIG_DISABLE_PCIEASPM that's under EXPERT, and people who 
> > really want to disable the code can do so.
> 
> I think the fact that the EXPERT didn't get removed in the above bug
> is a defintive bug.  But I'd go further and think the CONFIG_PCIEASPM
> option should be removed entirely.  There is absolutely no good reason
> to not build this small amount of code if PCIe support is enabled.

That might be a good solution.  I don't really want a situation where
leaving CONFIG_PCIEASPM unset is "known to break PME."

ASPM support has historically been fragile, and it's actually a
significant amount of code (~10KB on my x86), which is larger than
many other things for which we have config options.  But we do have
boot-time and run-time ways to disable it.

Another possibility might be to make it so Linux always advertises
ASPM support, but doesn't actually *do* anything unless
CONFIG_PCIEASPM is set.  I don't think advertising ASPM support means
the OS is required to change whatever configuration firmware did.

Bjorn