Re: [E1000-devel] igb driver with Intel Atom Bay Trail issue

[Andy Shevchenko <andriy.shevchenko@linux.intel.com>]

On Wed, Mar 27, 2019 at 05:02:31PM +0100, Hans de Goede wrote:
> Hi,
> 
> On 3/25/19 9:12 PM, Fujinaka, Todd wrote:
> > This is going to take a bit of time to see what we need to do. The attachments were stripped but I think just figuring out what they had to change in the Realtek driver will tell us what we need to know.
> 
> I'm not sure fixing this in the ethernet driver is the best way to go,
> the board in question is an embedded PC and I've also recently received
> a bug report about a similar problem with the consumer not requesting
> the pmc clocks causing a HSIC usb hub to not work.
> 
> I think that it might be better to restore the CLK_IS_CRITICAL workaround
> for this, but then only on select boards, based on DMI matching.
> 
> I've added a bunch of relevant people / lists to the Cc.
> 
> Andy, Stephen, what is your take on this ?

I'm afraid I forgot all details about that (semi-)famous issue. Though, looking
into your patch against r8169 and taking into account DT practice, it would be
not bad to fix a driver, we have by the way devm_clk_get_optional() now, so, it
would be not a big deal.

> I'm myself starting to believe the DMI based applying of the
> CLK_IS_CRITICAL workaround is the best solution here.

DMI quirk table tends to grow in mysterious ways. I would prefer in this case
logical solution — if platform has an optional clock, then use it.

> 
> Regards,
> 
> Hans
> 
> 
> > 
> > Todd Fujinaka
> > Software Application Engineer
> > Datacenter Engineering Group
> > Intel Corporation
> > todd.fujinaka@intel.com
> > 
> > 
> > -----Original Message-----
> > From: Semyon Verchenko [mailto:semverchenko@factor-ts.ru]
> > Sent: Monday, March 25, 2019 6:25 AM
> > To: hdegoede@redhat.com; Kirsher, Jeffrey T <jeffrey.t.kirsher@intel.com>; e1000-devel@lists.sf.net
> > Subject: [E1000-devel] igb driver with Intel Atom Bay Trail issue
> > 
> > Dear Linux/igb maintainers,
> > 
> > We've encountered problem with igb driver (both one that is distributed with Linux kernel and one which is downloadable from intel.com). After commit 648e921888ad ("clk: x86: Stop marking clocks as CLK_IS_CRITICAL") on machine with Intel(R) Atom(TM) CPU E3825 with 4 ethernet cards Intel
> > I211 only one of cards probed correctly, other ones fail with error -2 (error -5 with driver from intel.com). I've rebuilt kernel with commit 648e921888ad reverted and all 4 interfaces had to probe correctly.
> > Problem is reproducible at least with kernel 5.0.4 and kernels 4.14.y (actually firstly I've encountered this on 4.14.105 while it worked fine with 4.14.67 so I firstly started to build intermediate versions of kernel and found that problem started to appear on 4.14.77, but since it's reproducible in mainline kernel I think I should inform mainline kernel maintainers about this). Also it is reproducible with igb
> > 5.3.5.22 from intel.com. I'm attaching kernel logs (journalctl.bad is from stock kernel and journalctl.badfix is from stock kernel with commit 648e921888ad reversed), lspci -vnn output (same about file names) and /proc/cpuinfo. The system is Arch Linux with kernel replaced to stock one (it is not Arch Linux-related problem as it appeared in another system with 4.14 kernel). I think something like Hans de Goede's r8169 patches is required for igb (and maybe other Intel drivers), but I'm not so good with driver developing so it seems too hard for me.
> > 
> > Regards,
> > Semyon Verchenko
> > 

-- 
With Best Regards,
Andy Shevchenko