Build/qemu test results for v4.18-rc2

Build/qemu test results for v4.18-rc2

[Guenter Roeck]

Build results:
	total: 134 pass: 113 fail: 21
Failed builds:
	arm:allmodconfig
	i386:allyesconfig
	i386:allmodconfig
	mips:defconfig
	mips:allnoconfig
	mips:defconfig
	mips:allmodconfig
	mips:allnoconfig
	mips:bcm47xx_defconfig
	mips:bcm63xx_defconfig
	mips:nlm_xlp_defconfig
	mips:ath79_defconfig
	mips:ar7_defconfig
	mips:e55_defconfig
	mips:cavium_octeon_defconfig
	mips:malta_defconfig
	mips:rt305x_defconfig
	mips:defconfig
	nds32:defconfig
	nds32:allnoconfig
	sparc32:allmodconfig
Qemu test results:
	total: 153 pass: 140 fail: 13
Failed tests:
	mips:malta_defconfig:nosmp
	mips:malta_defconfig:smp
	mips64:malta_defconfig:nosmp
	mips64:malta_defconfig:smp
	mipsel:24Kf:malta_defconfig:nosmp:initrd
	mipsel:24Kf:malta_defconfig:smp:initrd
	mipsel:24Kf:malta_defconfig:smp:rootfs
	mipsel:mips32r6-generic:malta_32r6_defconfig:smp:rootfs
	mipsel64:malta_defconfig:nosmp:rootfs
	mipsel64:malta_defconfig:smp:initrd
	mipsel64:malta_defconfig:smp:rootfs
	mipsel64:fuloong2e_defconfig:fulong2e:rootfs
	mipsel64:64r6el_defconfig:boston:rootfs

Only the mips problem is new, introduced with commit784e0300fe9fe ("rseq: Avoid infinite
recursion when delivering SIGSEGV"). The other problems have all been previously reported.

Guenter

Re: Build/qemu test results for v4.18-rc2

[Linus Torvalds]

On Mon, Jun 25, 2018 at 12:25 AM Guenter Roeck <linux@roeck-us.net> wrote:
>
> Build results:
>         total: 134 pass: 113 fail: 21
> Failed builds:
>         i386:allyesconfig
>         i386:allmodconfig

These seem to be due to the trial code to embed user modules in the
netfilter modules to use bpf  to do netfilter (aka bpfilter). Right
now that doesn't cross-compile right.

I'm hoping the networking people get their trial balloon to work
right, but if worst comes to worst, it will just be disabled for 4.18.

In the meantime, I'm hoping it doesn't hold back any *other* testing.
I'm assuming the allyes/modconfig build failures don't actually matter
for the boot testing etc.

The mips problems I have no idea about.

             Linus

Re: Build/qemu test results for v4.18-rc2

[Thomas Gleixner]

On Mon, 25 Jun 2018, Linus Torvalds wrote:
> On Mon, Jun 25, 2018 at 12:25 AM Guenter Roeck <linux@roeck-us.net> wrote:
> >
> > Build results:
> >         total: 134 pass: 113 fail: 21
> > Failed builds:
> >         i386:allyesconfig
> >         i386:allmodconfig
> 
> These seem to be due to the trial code to embed user modules in the
> netfilter modules to use bpf  to do netfilter (aka bpfilter). Right
> now that doesn't cross-compile right.
> 
> I'm hoping the networking people get their trial balloon to work
> right, but if worst comes to worst, it will just be disabled for 4.18.
> 
> In the meantime, I'm hoping it doesn't hold back any *other* testing.
> I'm assuming the allyes/modconfig build failures don't actually matter
> for the boot testing etc.
> 
> The mips problems I have no idea about.

Fix is in the mips tree already.

Re: Build/qemu test results for v4.18-rc2

[Guenter Roeck]

On Mon, Jun 25, 2018 at 03:17:47PM +0800, Linus Torvalds wrote:
> On Mon, Jun 25, 2018 at 12:25 AM Guenter Roeck <linux@roeck-us.net> wrote:
> >
> > Build results:
> >         total: 134 pass: 113 fail: 21
> > Failed builds:
> >         i386:allyesconfig
> >         i386:allmodconfig
> 
> These seem to be due to the trial code to embed user modules in the
> netfilter modules to use bpf  to do netfilter (aka bpfilter). Right
> now that doesn't cross-compile right.
> 

Yes, looks like they tried again with c24fb5e68ebf ("bpfilter: fix user
mode helper cross compilation"), but it still doesn't work.

Build reference: v4.18-rc2-37-g6f0d349
gcc version: x86_64-linux-gcc.br_real (Buildroot 2017.02) 6.3.0

Building i386:defconfig ... passed
Building i386:allyesconfig ... failed

x86_64-linux-ld: i386:x86-64 architecture of input file
	`net/bpfilter/bpfilter_umh.o' is incompatible with i386 output

Guenter

Re: Build/qemu test results for v4.18-rc2

[Linus Torvalds]

On Mon, Jun 25, 2018 at 9:31 AM Guenter Roeck <linux@roeck-us.net> wrote:
>
> Yes, looks like they tried again with c24fb5e68ebf ("bpfilter: fix user
> mode helper cross compilation"), but it still doesn't work.

That was a different issue - an actual cross-build environment.

The problem on x86-64 cross-building to i386 is that the *compiler* is
the same, but the flags are not, and the games the Makefile plays
aren't sufficient. It plays with changing HOSTCC to CC in order to get
he right target compiler, but since the difference between i386 and
x86-64 is not the compiler but the "-m64" in the cflags, it doesn't
actually do what it intended to do.

We'll get it sorted, I think. The problem is that the current Makefile
infrastructure doesn't have any support for building a stand-alone
binary for the target, which is why the code plays games with HOSTCC
(to use the infrastructure that does binaries for the host, but using
the target compiler).

So it's all a bit hacky right now, and the hacks come back and bite us
for the x86-64/i386 case.

               Linus

Re: Build/qemu test results for v4.18-rc2

[Linus Torvalds]

On Tue, Jun 26, 2018 at 8:40 AM Linus Torvalds
<torvalds@linux-foundation.org> wrote:
>
> The problem on x86-64 cross-building to i386 is that the *compiler* is
> the same, but the flags are not, and the games the Makefile plays
> aren't sufficient. It plays with changing HOSTCC to CC in order to get
> he right target compiler, but since the difference between i386 and
> x86-64 is not the compiler but the "-m64" in the cflags, it doesn't
> actually do what it intended to do.
>
> We'll get it sorted, I think.

So right now the build fix is likely to be this:

    https://patchwork.ozlabs.org/patch/935246/

but note that it's literally just a build fix, and for a i386
cross-compile it embeds the wrong binary so it won't actually run
correctly. For this whole "test the concept" stage that's fine.

                Linus

Re: Build/qemu test results for v4.18-rc2

[Guenter Roeck]

On Wed, Jun 27, 2018 at 08:27:47AM -0700, Linus Torvalds wrote:
> On Tue, Jun 26, 2018 at 8:40 AM Linus Torvalds
> <torvalds@linux-foundation.org> wrote:
> >
> > The problem on x86-64 cross-building to i386 is that the *compiler* is
> > the same, but the flags are not, and the games the Makefile plays
> > aren't sufficient. It plays with changing HOSTCC to CC in order to get
> > he right target compiler, but since the difference between i386 and
> > x86-64 is not the compiler but the "-m64" in the cflags, it doesn't
> > actually do what it intended to do.
> >
> > We'll get it sorted, I think.
> 
> So right now the build fix is likely to be this:
> 
>     https://patchwork.ozlabs.org/patch/935246/
> 

I can confirm that the patch fixes the i386 cross-compile issue.
I just sent a Tested-by:.

Guenter

> but note that it's literally just a build fix, and for a i386
> cross-compile it embeds the wrong binary so it won't actually run
> correctly. For this whole "test the concept" stage that's fine.
> 
>                 Linus