This is the mail archive of the
ecos-discuss@sourceware.org
mailing list for the eCos project.
Massive ISR to DSR delay caused by free()
- From: "Peter Graf" <p dot graf at itknet dot de>
- To: <ecos-discuss at ecos dot sourceware dot org>
- Date: Tue, 03 Dec 2013 11:19:17 +0100
- Subject: Massive ISR to DSR delay caused by free()
- Authentication-results: sourceware.org; auth=none
Hi,
under rare circumstances, we experienced a massive timing overflow of a cyclical DSR. It turned out to be an issue with JFFS2, occurring after a system reboot if long files had been written to the filesystem before.
I found that the issue was only caused by JFFS2's
malloc-ecos.c: jffs2_free_full_dnode()
By using an oscilloscope, I could see that after the ISR was served, the DSR was delayed, just because free() was called by jffs2_free_full_dnode(). More precisely, the CPU time was spent in memalloc's
mvarimpl.inl: Cyg_Mempool_Variable_Implementation::insert_free_block()
At first I thought it was a locking issue within JFFS2. But in the end I could not see any scheduler locks which might have caused it.
I now have indications, that it could be a general issue with free(). A call to free() from even a low priority thread seems to delay DSR execution after the ISR!
The allocator is compiled to be thread-safe, but in my opinion it would be a major drawback if even a low priority thread could not call free() without delaying a critical DSR.
Can someone confirm this behaviour? Is it intended?
Many thanks,
Peter
--
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss