Re: linux-next: build warning after merge of the net-next tree

On 08/14/2018 12:05 AM, Masahiro Yamada wrote:

2018-07-20 8:19 GMT+09:00 Masahiro Yamada [off-list ref]:

quoted

2018-07-20 7:35 GMT+09:00 Guenter Roeck [off-list ref]:

quoted

On Fri, Jul 20, 2018 at 08:09:06AM +1000, Stephen Rothwell wrote:

quoted

Hi Guenter,

On Thu, 19 Jul 2018 06:49:01 -0700 Guenter Roeck [off-list ref] wrote:

quoted

On 07/18/2018 10:29 PM, Stephen Rothwell wrote:

quoted

On Wed, 18 Jul 2018 20:52:56 -0700 Guenter Roeck [off-list ref] wrote:

quoted

On 07/18/2018 07:04 PM, Stephen Rothwell wrote:

quoted

After merging the net-next tree, today's linux-next build (x86_64
allmodconfig) produced this warning:

*
* Restart config...
*
....

This is output by my "make allmodconfig" and only started after merging
the net-next tree today.  It has continued for further merges/builds.

I suspect commit

     1323061a018a ("net: phy: sfp: Add HWMON support for module sensors")

which added an "imply" clause.

I thought "imply" was better than "depends on HWMON || HWMON=n", but maybe
not. Is that a caveat when using "imply", and does it mean that "imply"
should better not be used ?

I don't know, sorry.  It was just my best guess from what I could see
had changed.

I wonder if it makes a difference that I am doing my "make
allmodconfig" on top of a previous "make allmodconfig" and some symbols
are marked as "NEW" (though they are not symbols related to the changes
that happened during the net-next tree merge)?

I tried to reproduce the problem, but I don't see the message.

What I do see, though, is that "make allmodconfig" on a clean tree,
followed by "make menuconfig", results in configuration changes.
Specifically,

  > CONFIG_ARC_EMAC_CORE=m
  > CONFIG_ARC_EMAC=m
  > CONFIG_EMAC_ROCKCHIP=m

is removed by menuconfig, and a large number of "# ... is not set"
configuration lines are added. Weird and bad, since several of the
disabled configurations _should_ be enabled by "make allmodconfig",
and a large number of hwmon drivers are affected. Bisect does point
to "net: phy: sfp: Add HWMON support for module sensors", meaning
"imply hwmon" does have severe side effects and can not be used.

I'll try to find a fix.

OK, my mistake, the "make allmodconfig" works, the following "make"
causes the config restart.  (I am actually doing cross builds and using
an external object directory, in case that matters.)

I removed the "imply HWMON" line added by the above commit and the
problem went away.  Also, using "depends on HWMON || HWMON=n" instead
of the imply fixes it.

Yes, replacing imply with the dependency is what I did in the fixup patch.
Sorry, I should have copied you: https://patchwork.kernel.org/patch/10534925/

It is a bit different - imply was supposed to enforce HWMON={y,n} if SFP=y,
and the depends line enforces SFP={n,m} if HWMON=m. I have no idea why
imply doesn't work, but I think I'll stay away from it in the future.

Guenter

Hmm, this could be a Kconfig bug.

I will take a look.

Today, I took a look at it.

The cause of the problem was the circular dependency.

Somehow, 'imply' is not checked in the circular dependency.

So, I wrote patches to report this.
https://patchwork.kernel.org/patch/10565061/
https://patchwork.kernel.org/patch/10565063/


If you apply those two patches on top of commit
1323061a018a ("net: phy: sfp: Add HWMON support for module sensors")
It is reported in allmodconfig stage, like this:


masahiro@grover:~/ref/linux-next$ make allmodconfig
   HOSTCC  scripts/kconfig/conf.o
   YACC    scripts/kconfig/zconf.tab.c
   HOSTCC  scripts/kconfig/zconf.tab.o
   HOSTLD  scripts/kconfig/conf
scripts/kconfig/conf  --allmodconfig Kconfig
drivers/of/Kconfig:68:error: recursive dependency detected!
drivers/of/Kconfig:68: symbol OF_IRQ depends on IRQ_DOMAIN
kernel/irq/Kconfig:63: symbol IRQ_DOMAIN is selected by REGMAP
drivers/base/regmap/Kconfig:6: symbol REGMAP is selected by SENSORS_ASPEED
drivers/hwmon/Kconfig:352: symbol SENSORS_ASPEED depends on HWMON
drivers/hwmon/Kconfig:5: symbol HWMON is implied by SFP
drivers/net/phy/Kconfig:214: symbol SFP depends on PHYLIB
drivers/net/phy/Kconfig:181: symbol PHYLIB is selected by ARC_EMAC_CORE
drivers/net/ethernet/arc/Kconfig:18: symbol ARC_EMAC_CORE is selected
by ARC_EMAC
drivers/net/ethernet/arc/Kconfig:24: symbol ARC_EMAC depends on OF_IRQ
For a resolution refer to Documentation/kbuild/kconfig-language.txt
subsection "Kconfig recursive dependency limitations"

Interesting; I thought that "implied" does not really select the symbol.
That means 'imply' just doesn't work for HWMON, and we'll have to stick
with the old "depends on HWMON || HWMON=n".

Thanks,
Guenter