2010-12-15 04:59:21 Xenomai warning in 2010R1: "kernel is anterior to 2.5.2"?
Kolja Waschk (GERMANY)
Message: 96762
Hi, I started to use blackfin-linux-dist 2010R1-RC5 as a base for a project which previously was based on 2009R1.1. Some of the applications are utilizing Xenomai, and all (including switchtest etc.) now emit a warning:
Xenomai: WARNING, this version of Xenomai kernel is anterior to 2.5.2.
It can cause memory corruption on thread termination.
Upgrade is recommended.
However, the Xenomai kernel is 2.5.3:
/ # cat /proc/xenomai/version
2.5.3
I found that above warning is given dependent on the presence of XNVDSO_FEAT_DROP_U_MODE (i.e. the warning is made if the feature is not present):
xenomai-2.5.3/src/skins/common/sem_heap.c:
if (!xnvdso_test_feature(XNVDSO_FEAT_DROP_U_MODE))
But as far as I can see in the sources in the linux-2.6.x subtree, the feature should be present. Maybe I've made some configuration that implicitly disables such feature? Any other idea where I could look further?
My build process is as follows due to the way the Xenomai patch is applied to the kernel:
untar blackin-linux-dist*
make MyVendor/Product_default
cp MyVendor/Product/config.linux-2.6.x linux-2.6.x/.config
make oldconfig
make
The steps 3..5 are necessary because the XENOMAI settings from the vendor config.linux-2.6.x are ignored at first, because the kernel (at the time when it is configured) doesn't have Xenomai patch applied. I'm not sure if this might contribute to the "problem". I don't even know for sure whether it is actually a problem or just cosmetic.
Thanks for hints
Kolja
QuoteReplyEditDelete
2010-12-15 05:28:26 Re: Xenomai warning in 2010R1: "kernel is anterior to 2.5.2"?
Aaron Wu (CHINA)
Message: 96763
Xenomai related test get passed in this release, so guess you can ignore that.
QuoteReplyEditDelete
2010-12-15 06:26:09 Re: Xenomai warning in 2010R1: "kernel is anterior to 2.5.2"?
Kolja Waschk (GERMANY)
Message: 96764
Hi,
So I may assume that during your tests you have seen the very same warning? I haven't found any tracker item regarding this behaviour? Warnings printed by library code usually have important reasoning.
The warnings are about memory corruption after thread termination. That's not something that can be detected easily by tests, especially if the corruption is caused by OS and not application code. However, it may be harmful for the system integrity especially on nommu systems.
So I'd rather like to find the actual cause why the warning is printed in order to decide whether I may ignore it. I'll post here or report in a tracker item if I find more info.
Thanks
Kolja
QuoteReplyEditDelete
2010-12-15 06:57:02 Re: Xenomai warning in 2010R1: "kernel is anterior to 2.5.2"?
Kolja Waschk (GERMANY)
Message: 96766
Quick note: It seems the features are misinterpreted in the library code because the value is read from corrupted memory or from a wrong location :/ - maybe a later Xenomai code fixes this, at least it has changes that directly affect this part...
QuoteReplyEditDelete
2010-12-15 08:25:46 Re: Xenomai warning in 2010R1: "kernel is anterior to 2.5.2"?
Kolja Waschk (GERMANY)
Message: 96767
Using Xenomai 2.5.5.2 fixes it. Granted, I also have no specific test but the warning is gone and there are some sections in the affectd code (sem_heap.c) that specifically address nommu targets...
QuoteReplyEditDelete
2010-12-15 08:53:50 Re: Xenomai warning in 2010R1: "kernel is anterior to 2.5.2"?
Kolja Waschk (GERMANY)
Message: 96768
Correction: With 2.5.5.2 it also happens, but rarely (e.g. once when calling cyclictest)
QuoteReplyEditDelete
2010-12-15 11:57:11 Re: Xenomai warning in 2010R1: "kernel is anterior to 2.5.2"?
Kolja Waschk (GERMANY)
Message: 96769
Filed as Tracker Item 6408