Message-ID: <TYAPR01MB45447DFE9E81D77CA867DEC8D8C20@TYAPR01MB4544.jpnprd01.prod.outlook.com>
Date:   Mon, 6 Apr 2020 08:43:10 +0000
From:   Yoshihiro Shimoda <yoshihiro.shimoda.uh@...esas.com>
To:     John Stultz <john.stultz@...aro.org>,
        Geert Uytterhoeven <geert@...ux-m68k.org>
CC:     "gregkh@...uxfoundation.org" <gregkh@...uxfoundation.org>,
        "rafael@...nel.org" <rafael@...nel.org>,
        "iommu@...ts.linux-foundation.org" <iommu@...ts.linux-foundation.org>,
        LKML <linux-kernel@...r.kernel.org>,
        netdev <netdev@...r.kernel.org>
Subject: RE: How to fix WARN from drivers/base/dd.c in next-20200401 if
 CONFIG_MODULES=y?

Hi John, Geert,

> From: John Stultz, Sent: Saturday, April 4, 2020 1:19 PM
> 
> On Fri, Apr 3, 2020 at 4:47 AM Geert Uytterhoeven <geert@...ux-m68k.org> wrote:
> > On Thu, Apr 2, 2020 at 7:27 PM John Stultz <john.stultz@...aro.org> wrote:
> > > On Thu, Apr 2, 2020 at 3:17 AM Yoshihiro Shimoda
> > > <yoshihiro.shimoda.uh@...esas.com> wrote:
> > > >
> > > > I found an issue after applied the following patches:
> > > > ---
> > > > 64c775f driver core: Rename deferred_probe_timeout and make it global
> > > > 0e9f8d0 driver core: Remove driver_deferred_probe_check_state_continue()
> > > > bec6c0e pinctrl: Remove use of driver_deferred_probe_check_state_continue()
> > > > e2cec7d driver core: Set deferred_probe_timeout to a longer default if CONFIG_MODULES is set
> >
> > Note that just setting deferred_probe_timeout = -1 like for the
> > CONFIG_MODULES=n case doesn't help.
> 
> Yea. I can see why in that case, as we're checking
> !IS_ENABLED(CONFIG_MODULES) directly in
> driver_deferred_probe_check_state.
> 
> I guess we could switch that to checking
> (driver_deferred_probe_timeout == -1) which would have the same logic
> and at least make it consistent if someone specifies -1 on the command
> line (since now it will effectively have it EPROBE_DEFER forever in
> that case). But also having a timeout=infinity could be useful if
> folks don't want the deferring to time out.  Maybe in the !modules
> case setting it to =0 would be the most clear.
> 
> But that's sort of a further cleanup. I'm still more worried about the
> NFS failure below.
> 
> 
> > > Hey,
> > >   Terribly sorry for the trouble. So as Robin mentioned I have a patch
> > > to remove the WARN messages, but I'm a bit more concerned about why
> > > after the 30 second delay, the ethernet driver loads:
> > >   [   36.218666] ravb e6800000.ethernet eth0: Base address at
> > > 0xe6800000, 2e:09:0a:02:eb:2d, IRQ 117.
> > > but NFS fails.
> > >
> > > Is it just that the 30 second delay is too long and NFS gives up?
> >
> > I added some debug code to mount_nfs_root(), which shows that the first
> > 3 tries happen before ravb is instantiated, and the last 3 tries happen
> > after.  So NFS root should work, if the network works.
> >
> > However, it seems the Ethernet PHY is never initialized, hence the link
> > never becomes ready.  Dmesg before/after:
> >
> >      ravb e6800000.ethernet eth0: Base address at 0xe6800000,
> > 2e:09:0a:02:ea:ff, IRQ 108.
> >
> > Good.
> >
> >      ...
> >     -gpio_rcar e6052000.gpio: sense irq = 11, type = 8
> >
> > This is the GPIO the PHY IRQ is connected to.
> > Note that that GPIO controller has been instantiated before.
> >
> >      ...
> >     -Micrel KSZ9031 Gigabit PHY e6800000.ethernet-ffffffff:00:
> > attached PHY driver [Micrel KSZ9031 Gigabit PHY]
> > (mii_bus:phy_addr=e6800000.ethernet-ffffffff:00, irq=197)
> >      ...
> >     -ravb e6800000.ethernet eth0: Link is Up - 1Gbps/Full - flow control off
> >
> > Oops.
> >
> >     -Sending DHCP requests .., OK
> >     -IP-Config: Got DHCP answer from ...
> >      ...
> >     +VFS: Unable to mount root fs via NFS, trying floppy.
> >     +VFS: Cannot open root device "nfs" or unknown-block(2,0): error -6
> >
> > > Does booting with deferred_probe_timeout=0 work?
> >
> > It does, as now everything using optional links (DMA and IOMMU) is now
> > instantiated on first try.
> 
> Thanks so much for helping clarify this!
> 
> So it's at least good to hear that booting with
> deferred_probe_timeout=0 is working!  But I'm bummed the NFS (or as
> you pointed out in your later mail,  ip_auto_config) falls over
> because the network isn't immediately there.
> 
> Looking a little closer at the ip_auto_config() code, I think the
> issue may be that wait_for_device_probe() is effectively returning too
> early, since the probe_defer_timeout is still active? I need to dig a
> bit more on that code, on Monday, as I don't fully understand it yet.

I think so. I also investigated this issue more and then the following
patch seems to be related because return value is changed a bit.

c8c43ce driver core: Fix driver_deferred_probe_check_state() logic

# By the way, this is other topic though, IIUC we should revise
# the deferred_probe_timeout= in Documentation/admin-guide/kernel-parameters.txt
# for the commit c8c43ce. Especially " A timeout of 0 will timeout at the end of initcalls."
# doesn't match after we applied the commit.

I'm guessing we should add the following flush_work for deferred_probe_timeout_work().
# Sorry, I didn't test this for some reasons yet though...

+       /* wait for the deferred probe timeout workqueue to finish */
+       if (driver_deferred_probe_timeout > 0)
+               flush_work(&deferred_probe_timeout_work);

> If I can't find a way to address that, I think the best course will be
> to set the driver_deferred_probe_timeout value to default to 0
> regardless of the value of CONFIG_MODULES, so we don't cause any
> apparent regression from previous behavior. That will also sort out
> the less intuitive = -1 initialization in the non-modules case.
> 
> In any case, I'll try to have a patch to send out on Monday.

Thanks!

Best regards,
Yoshihiro Shimoda

> thanks
> -john

Powered by blists - more mailing lists