Re: nsdeps not working on modules in 5.4-rc1

From: Matthias Maennich
Date: Fri Oct 04 2019 - 06:40:17 EST

Next message: Dmitry Goldin: "[PATCH v2] kheaders: making headers archive reproducible"
Previous message: Adam Ford: "Re: DM3730 Bluetooth Performance differences between SERIAL_8250_OMAP vs SERIAL_OMAP"
In reply to: Steve French: "Re: nsdeps not working on modules in 5.4-rc1"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Hi Steve,

On Fri, Oct 04, 2019 at 12:01:20AM -0500, Steve French wrote:

ok - so to sum up, it sounds like you are saying the 350 false
positives (see attached file) that happen on building cifs.ko .will be
fixed by a
future change to modpost? This is a typical module build ...

Yes, the reason for you to see these false positive warnings is a bug in
modpost. The patch series we were referring to addresses this problem
(and some others). I do not believe you need to run nsdeps to fixup
anything in fs/cifs at this time.

download and install current kernel package (in this case Ubuntu
5.4-rc1). which saves huge amount of build time, then build just the
module of interest (in my case cifs.ko)

cd fs/cifs
make C=1 -C /usr/src/linux-headers-`uname -r` M=`pwd` modules

If nsdeps is not needed to fixup some namespace issue then shouldn't
be a problem, just trying to avoid the distraction of 300+
warning messages every time I build just this one module. Is there a
workaround?

Sorry for the noise this creates. The only known workaround is to
locally apply the complete patch series.

Cheers,
Matthias

On Thu, Oct 3, 2019 at 11:51 PM Masahiro Yamada
<yamada.masahiro@xxxxxxxxxxxxx> wrote:

Hi Steve,

On Fri, Oct 4, 2019 at 1:28 PM Steve French <smfrench@xxxxxxxxx> wrote:
>
> On Thu, Oct 3, 2019 at 10:41 PM Masahiro Yamada
> <yamada.masahiro@xxxxxxxxxxxxx> wrote:
> >
> > Hi Steve,
> >
> > On Fri, Oct 4, 2019 at 1:07 AM Steve French <smfrench@xxxxxxxxx> wrote:
> > >
> > > On Thu, Oct 3, 2019 at 10:24 AM Masahiro Yamada
> > > <yamada.masahiro@xxxxxxxxxxxxx> wrote:
> > > >
> > > > Hi Steve,
> > > >
> > > > On Fri, Oct 4, 2019 at 12:15 AM Steve French <smfrench@xxxxxxxxx> wrote:
> > > > >
> > > > > On Thu, Oct 3, 2019 at 5:43 AM Matthias Maennich <maennich@xxxxxxxxxx> wrote:
> > > > > >
> > > > > > Hi Steve!
> > > > > >
> > > > > > On Wed, Oct 02, 2019 at 06:54:26PM -0500, Steve French wrote:
> > > > > > >And running the build differently, from the root of the git tree
> > > > > > >(5.4-rc1) rather than using the Ubuntu 5.4-rc1 headers also fails
> > > > > > >
> > > > > > >e.g. "make M=fs/cifs modules nsdeps"
> > > > > > >
> > > > > > >...
> > > > > > > LD [M] fs/cifs/cifs.o
> > > > > > > Building modules, stage 2.
> > > > > > > MODPOST 1 modules
> > > > > > >WARNING: module cifs uses symbol sigprocmask from namespace
> > > > > > >_fs/cifs/cache.o), but does not import it.
> > > > > > >...
> > > > > > >WARNING: module cifs uses symbol posix_test_lock from namespace
> > > > > > >cifs/cache.o), but does not import it.
> > > > > > > CC [M] fs/cifs/cifs.mod.o
> > > > > > > LD [M] fs/cifs/cifs.ko
> > > > > > > Building modules, stage 2.
> > > > > > > MODPOST 1 modules
> > > > > > >./scripts/nsdeps: 34: local: ./fs/cifs/cifsfs.c: bad variable name
> > > > > > >make: *** [Makefile:1710: nsdeps] Error 2
> > > > > >
> > > > > > Thanks for reporting this. It appears to me you hit a bug that was
> > > > > > recently discovered: when building with `make M=some/subdirectory`,
> > > > > > modpost is misbehaving. Can you try whether this patch series solves
> > > > > > your problems:
> > > > > > https://lore.kernel.org/lkml/20191003075826.7478-1-yamada.masahiro@xxxxxxxxxxxxx/
> > > > > > In particular patch 2/6 out of the series.
> > > > > >
> > > > > > Cheers,
> > > > > > Matthias
> > > > >
> > > > >
> > > > > Applying just patch 2 and doing "make" from the root of the git tree
> > > > > (5.4-rc1), at the tail end of the build I got
> > > > >
> > > > > ...
> > > > > Kernel: arch/x86/boot/bzImage is ready (#87)
> > > > > Building modules, stage 2.
> > > > > MODPOST 5340 modules
> > > > > free(): invalid pointer
> > > > > Aborted (core dumped)
> > > >
> > > >
> > > > Right.
> > > >
> > > > Since 2/6 depends on 1/6,
> > > > applying only the second one does not work.
> > >
> > > Applying both 1 and 2 I get the following error doing make (although
> > > it makes it a long way into the build)
> > >
> > > <snip>
> > > WARNING: drivers/usb/storage/usb-storage: 'USB_STORAGE' exported
> > > twice. Previous export was in drivers/usb/storage/usb-storage.ko
> > > ERROR: "usb_stor_set_xfer_buf" [drivers/usb/storage/ums-usbat.ko] undefined!
> > > ERROR: "usb_stor_access_xfer_buf" [drivers/usb/storage/ums-usbat.ko] undefined!
> > > ERROR: "usb_stor_post_reset" [drivers/usb/storage/ums-usbat.ko] undefined!
> > > ERROR: "usb_stor_disconnect" [drivers/usb/storage/ums-usbat.ko] undefined!
> > > <snip>
> > > ERROR: "usb_stor_adjust_quirks" [drivers/usb/storage/uas.ko] undefined!
> > > ERROR: "usb_stor_sense_invalidCDB" [drivers/usb/storage/uas.ko] undefined!
> > > WARNING: "USB_STORAGE" [drivers/usb/storage/usb-storage] is a static
> > > EXPORT_SYMBOL_GPL
> > > make[1]: *** [scripts/Makefile.modpost:94: __modpost] Error 1
> > > make: *** [Makefile:1303: modules] Error 2
> >
> >
> > Hmm, I do not see those error.
> > I was able to build the kernel successfully.
> > (I asked the 0-day bot to test whole of my patch set
> > in case I am missing something.)
> >
> >
> > Could you share the steps to reproduce the errors and your .config file?
>
> From the root of git tree - at exactly 5.4-rc1
>
> ~/cifs-2.6$ make nsdeps
> CALL scripts/checksyscalls.sh
> CALL scripts/atomic/check-atomics.sh
> DESCEND objtool
> CHK include/generated/compile.h
> CHK kernel/kheaders_data.tar.xz
> Building modules, stage 2.
> MODPOST 5340 modules
> Building modules, stage 2.
> MODPOST 5340 modules
> ./scripts/nsdeps: 34: local: ./fs/cifs/cifsfs.c: bad variable name
> make: *** [Makefile:1710: nsdeps] Error 2
>
> I get the same error doing "rm fs/cifs/*.o" and repeating the "make
> nsdeps" command
>
> I will send you the .config

You need to clarify your problem.

In the first post from you, you pointed out
"hundreds of new warnings introduced by namespaces in 5.4-rc1 when
building my module"

So, 1/6 and 2/6 should address that problem.
https://lore.kernel.org/patchwork/patch/1133628/
https://lore.kernel.org/patchwork/patch/1133626/

Then, in the previous email ("Applying both 1 and 2 I get the
following error doing make")
it looks like you were talking about in-kernel building
instead of the external module.

Then, in this email, you are talking about "make nsdeps".
"make nsdeps" obviously does not support M=.

I am afraid you are mixing up different issues,
which is so confusing. (and I am afraid you were too confused)

Currently, the namespace is used only by USB_STORAGE.
So, it should not be a problem for your module.

To sum up, you do not need to run nsdeps at all.
The hundreds of false-positive warnings came from the modpost bug,
and should be fixed soon.

--
Best Regards
Masahiro Yamada

--
Thanks,

Steve

Next message: Dmitry Goldin: "[PATCH v2] kheaders: making headers archive reproducible"
Previous message: Adam Ford: "Re: DM3730 Bluetooth Performance differences between SERIAL_8250_OMAP vs SERIAL_OMAP"
In reply to: Steve French: "Re: nsdeps not working on modules in 5.4-rc1"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]