Re: Detecting PREEMPT_RT
From: John Kacur <jkacur@redhat.com>
Date: 2013-01-21 11:36:19
On Sun, Jan 20, 2013 at 9:18 PM, Carsten Emde [off-list ref] wrote:
John,quoted
quoted
quoted
quoted
Unfortunately, this wiki article is old and unmaintained. You're probably right to be a bit suspicious when checking for the "-rt" suffix of the kernel release (uname -r). But looking for the occurrence of "PREEMPT RT" in the kernel version (uname -v) should work. An application may use the system call uname() and check the version element of the utsname structure.the utsname.version string match is fine, however:quoted
In addition you may want to make sure the system has high-resolutiontimers. If so, the timers in /proc/timer_list have ".resolution: 1 nsecs". An application may use the function check_timer() from cyclictest for this purpose:quoted
static int check_timer(void) { struct timespec ts; if (clock_getres(CLOCK_MONOTONIC, &ts)) return 1; return (ts.tv_sec != 0 || ts.tv_nsec != 1); }Just tried this on a generic kernel (2.6.32-45-generic #102-Ubuntu SMP Wed Jan 2 21:53:06 UTC 2013 i686 GNU/Linux) and I get ts.tv_sec == 0 and ts.tv_nsec == 1, so it looks this cant be used to tell an RT_PREEMPT from a vanilla kernelI meant boolean AND when I said "In addition".Oh, I interpreted the timer check not as a PREEMPT_RT check, but a separate check for timers, which we must confirm are available before driving heavy machinery. So, if utsname.version contains "PREEMPT RT", then load the preempt-rt module. Once the module is loaded, from an init() function, run sanity checks, including the clock_getres check above. Is that about right?Yes. Sounds good to me.
Note we added a patch, called sysfs-realtime-entry.patch, but I'm not sure how far back it goes, at least it's in 3.x-rt kernels. This is the text from the patch "Add a /sys/kernel entry to indicate that the kernel is a realtime kernel. Clark says that he needs this for udev rules, udev needs to evaluate if its a PREEMPT_RT kernel a few thousand times and parsing uname output is too slow or so. Are there better solutions? Should it exist and return 0 on !-rt?" To use it, you do something like test -e /sys/kernel/realtime If you need it in an older kernel that doesn't have it, it should port trivially. Thanks