Re: [Xenomai] Question about kernel warning (Yocto/meta-linux + xenomai homebrew)

[Per Oberg <pero@wolfram.com>]

> On 2018-08-17 10:57, Per Oberg wrote:
> > Hi there

> > could someone please help me understand what this warning is:


> > [ 2932.508097] ------------[ cut here ]------------
>> [ 2932.514724] WARNING: CPU: 0 PID: 20645 at
> > /usr/src/kernel/arch/x86/include/asm/fpu/internal.h:538 fpu__drop+0x19e/0x200
> > [ 2932.529155] Modules linked in:
> > [ 2932.533513] CPU: 0 PID: 20645 Comm: cat Not tainted 4.9.90-xeno-cobolt #1
>> [ 2932.542067] Hardware name: Hewlett-Packard HP Z620 Workstation/158A, BIOS J61
> > v03.65 12/19/2013
> > [ 2932.552955] I-pipe domain: Linux
>> [ 2932.557675] ffffc90007f37dc8 ffffffff8143da78 0000000000000000
> > 0000000000000000
>> [ 2932.566582] ffffffff81ba4cc0 ffffc90007f37e08 ffffffff810785c1
> > 0000021a00000000
>> [ 2932.576484] 0000000000025dc0 ffff88041a1d5780 0000000000000246
> > 0000000000000000
> > [ 2932.585818] Call Trace:
> > [ 2932.590167] [<ffffffff8143da78>] dump_stack+0xbf/0xe7
> > [ 2932.593148] [<ffffffff810785c1>] __warn+0xe1/0x100
> > [ 2932.595426] [<ffffffff810786ad>] warn_slowpath_null+0x1d/0x20
> > [ 2932.597715] [<ffffffff8102754e>] fpu__drop+0x19e/0x200
> > [ 2932.600849] [<ffffffff81025a15>] exit_thread+0xa5/0xc0
> > [ 2932.603976] [<ffffffff8107bd7d>] do_exit+0x2bd/0xb40
> > [ 2932.606097] [<ffffffff81045605>] ? __do_page_fault+0x285/0x550
> > [ 2932.608534] [<ffffffff8107c683>] do_group_exit+0x43/0xb0
> > [ 2932.610757] [<ffffffff8107c704>] SyS_exit_group+0x14/0x20
> > [ 2932.613920] [<ffffffff81001c92>] do_syscall_64+0x82/0xf0
> > [ 2932.617071] [<ffffffff8190903e>] entry_SYSCALL_64_after_swapgs+0x58/0xc6
> > [ 2932.620705] ---[ end trace 8ef0d875e55c5ad1 ]---


> > At the time of running I was doing
> >> dohell 5&
> > followed by
> >> latency

> > Long story:
> > Xenomai version is 3.0.7,
> > Kernel is 4.9.90-ish

>> The kernel is a x86-64 yocto/meta-intel kernel patched with xenomai patches.
>> Unfortunately they didn't have their branch merged at Linux 4.9.90 so I had to
>> merge their mainline-linux branch into their project-branch myself, but there
>> were no changes from their part compared to their previous "release-tag" and
>> the merge was mostly ok except for some macros in intel hardware detection code
> > for different sound chips.

>> Also, patching the xenomai-patches (4.9.90-x86-6 something..) went ok, (except
> > for som stupid whitespace-stuff in a mmu-function).

>> I will try with another kernel but I was having some troubles with module
> > configuration since the default does not write any modules to the flash.

> I would indeed be helpful if you can reproduce the issue with the
> ipipe-4.9 kernel we provide. The yocto-linux kernel is unfortunately not
> in a good shape /wrt will-never-go-upstream patches (like yocto is in
> general), though those patches should not touch core infrastructure
> (AFAIK). What is the added-value of that kernel you would need?
> Jan
> --
> Siemens AG, Corporate Technology, CT RDA IOT SES-DE
> Corporate Competence Center Embedded Linux

I have been running xenomai-kernel patched from mainline for a few hours now and it does
not have the issue as it seems. The problem seemed to trigger quite easily before so 
I'm almost confident it's gone. 

About added value I've never delved deep enough into the  world of mainline linux vs distribution 
specific kernels to have an informed opinion. For my first attempts I ran into a few kernel panics because 
of missing drivers and I figured that the meta-intel and yocto defaults would give a better starting point. 
Now that I have a working configuration I am considering dropping their kernel in favor of a clean one. 

Thanks
Per Öberg