Hi all, I try to get PREEMP_RT running on an Olimex SAM9-L9260 board. On this board there is an AT91SAM9260 processor (ARM926EJ-S). Unfortunately more or less nothing works out of the box, whenever I try to use RT. I even have problems providing some useful data here (I'll try to explain why). I started with vanilla linux-2.6.31, no Atmel patch. Kernel .config was reduced to a minium. Kernel boots without a problem. When I enable debugging options and some tests like "Locking API Testsuite" I have no problems as well. When I apply patch-2.6.31-rt11 but do not enable full RT the kernel boots. When I try to get cyclictest (cyclictest -p99 -D5 -q -n) running I often get some unexpected results which are somewhat hard to describe. When I run cyclictest without any other process running (only /bin/ash from busybox) Max is often smaller than 1000 us. Sometimes I get 50000 us (no RT yet, so that is ok). When I start to ping -f the board, Max, Avg and Min start to increase, even if I stop the ping command, all three values are constantly increasing. In the given example above it is not uncommon that all three values are above 4'500'000. The same happens if I start a telnetd daemon in addition and login on the board. The fancy thing here that I can open new sessions and type some commands, therefore the system is still responsive. Without using the "-n" options results are often random numbers. Whenever I switch to RT (still linux-2.6.31) the kernel boots until init should be called. It is the same problem as here: http://article.gmane.org/gmane.linux.rt.user/5108. I switched back to linux-2.6.29.6 then. Main reason was that linuxlink.timesys.com offers a RT linux for AT91SAM9260 (an indication that should boot at least). Vanilla 2.6.29.6 boots fine. When I apply patch-2.6.29.6-rt23 without enabling RT, I get the same odd behavior with cyclictest as described above. When I enable RT, everything works more or less (tons of kernel Oops, often with at least one call to JFFS2 in the backtrace). Without any load, Max from cyclictest is below 500 us. If I ping -f the board Max goes up to ca. 8000 us. When I compare this results with: http://rt.wiki.kernel.org/index.php/CONFIG_PREEMPT_RT_Patch it should be possible to get worst case latency below 150 us (for this kind of processor). The problem with this page is that there are no details, about the test environment. Was it measured with cyclictest? Was there anything running in the background? Any testsuite that produces any load? Even details like, how much influence the gcc version has, would be interesting. It would be nice if anyone can post some details about their environment (especially ARM and PPC environment, PPC is missing in the list, whatever the reason). I thought then, ok let's track the source of the problem. I was reading http://www.osadl.org/Realtime-Preempt-Kernel.kernel-rt.0.html then. After enabling Function Tracing in the kernel, i was not able to compile the kernel anymore. I'm using the latest Code Sourcery GCC for ARM. I found a fix and if anyone is interested I'll post it here. Second problem is that it seems that cyclictest does not exactly works together with the current layout of debugfs (looks like some virtual files were moved around). Third problem ist when I try to interpret the output from debugfs I get thousands of lines prefixed with the exact same time. It seems impossible for me to get any idea, where the problem is. I've to admit that currently have so many problems that I don't now where I should start (write docs, fix tools, try to get more info about the Oops I get, get into the problem that somehow the results from cyclictest are sometimes totally wrong, ...). Thanks for any clarification, hint, or detailed description of a working ARM setup. Azraiyl -- To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
