From mboxrd@z Thu Jan 1 00:00:00 1970 From: Andy Shevchenko Subject: Re: r8169 broken when enabling the Atom PMC platform clocks Date: Mon, 10 Jul 2017 17:20:43 +0300 Message-ID: <1499696443.22624.327.camel@linux.intel.com> References: Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Return-path: Received: from mga11.intel.com ([192.55.52.93]:20658 "EHLO mga11.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932284AbdGJOUs (ORCPT ); Mon, 10 Jul 2017 10:20:48 -0400 In-Reply-To: Sender: platform-driver-x86-owner@vger.kernel.org List-ID: To: Carlo Caione , Platform Driver Cc: Darren Hart , irina.tirdea@intel.com, Pierre-Louis Bossart , Stephen Boyd , nic_swsd@realtek.com, Linux Upstreaming Team On Mon, 2017-07-10 at 16:15 +0200, Carlo Caione wrote: > Hi, > We are working on an Asus Z550M shipping a baytrail processor. From > the 4.11 kernel we noticed that the system is not bootable anymore > since it hangs during boot when probing the r8169 driver, not even a > trace is available. > > We bisected this problem down to commit 282a4e4 ("platform/x86: Enable > Atom PMC platform clocks"). > > We suspected that the problem is that one of the PMC clocks is being > used by the Ethernet board as XTAL clock and since it is not > explicitly claimed by the driver, it is gated at boot by the clock > framework, causing the system to hang. > > We have a quirk downstream in place where we basically modified the > r8169 driver to claim the 25MHz pmc_plt_clk_4 clock, and this seems to > work fine, but we really want to find a more upstreamable and > definitive solution. Can you copy in-place the hack patch you have? > The best solution would probably be avoiding to gate the clocks at all > when booting if these are being already used by the firmware, but IIUC > this information is not always available in the enable clock register. Yes, sounds sane. > Any idea how to approach this issue? I guess in the future we will see > more platforms needing a quirk like this because of fancy clock > routings. As simplest way we might do DMI matching, though I prefer to avoid as much as possible loading kernel by quirks. Pierre? Darren, it seems not the first report regarding this series. Like a quick fix we perhaps need to revert enabling patch and propagate to stable. Opinions? -- Andy Shevchenko Intel Finland Oy