[ltt-dev] User-space tracing on ARM?

Grant Edwards grant.b.edwards at gmail.com
Wed Oct 27 10:28:55 EDT 2010


On 2010-10-26, David Goulet <david.goulet at polymtl.ca> wrote:

> UST is "in theory" supported for ARM but hasn't been tested. However,
> two points to consider. First, it's not possible to synchronize LTTng
> traces with those of UST since the time source is not the same.

That's good to know.

> Second, on ARM architecture, clock_gettime, which is use in UST for
> the timestamp of each event, is a syscall and not a VDSO like on x86.
> This brings a *huge* impact on the tracing path.

I was wondering about that.  I've also been thinking about writing my
own simple event logging facility, and getting a timestamp value
without a system call is an unsolved piece of the puzzle.

> Having a "monotonic tracing clock source" for the Linux kernel and
> exported to userspace as a VDSO is an ongoing effort here :)
>
> For liburcu, ARM is supported. (See urcu/arch_* in liburcu tree for a
> complete list of all supported architecture)

I see that RCU only supports a single writer but multiple readers.
Does that mean you have to hve a separate log for every
process/thread?

> So, if you test UST on ARM, please post the results on ltt-dev
> mailing list and feel free to report any bugs.

-- 
Grant Edwards               grant.b.edwards        Yow! I joined scientology
                                  at               at a garage sale!!
                              gmail.com            





More information about the lttng-dev mailing list