Backporting LTTng 0.6.x to older kernels
Mathieu Desnoyers <[email protected]>
| Newsgroups | gmane.linux.kernel.tracing |
|---|---|
| Message-ID | <20061010131122.GB6200@Krystal> |
Hi bob, (I have CC'd ltt-dev as the information here can be useful to others) LTTng 0.6.x has all the latest bits. The only things that are different from the last versions is that it uses the marker mechanism which has not had all the testing LTTng 0.5.x had. It also uses debugfs and relay instead of relayfs, which makes my patch much less intrusive. The good news is : LTTng 0.6.x, by using the markers, makes backports and forward ports much easier than the previous 0.5.x. There is no more #include for each kernel file and no more code added at the assembly level. Syscalls are now traced with do_syscall_trace (or equivalent). It has been stable for me. The most problems I see comes from instrumentation of new 2.6.18 features like dynamic ticks which will require some changes on architecture without timestamp counters. The quickstart guide on the website is updated for LTTng 0.6.x. For the kernel/relay.c, include/linux/relay.h and fs/debugfs, include/linux/debugfs.h, you will probably have to copy the whole directories from a recent 2.6.17 or 2.6.18 to the older 2.6.9 kernel. Mathieu * Robert Wisniewski ([email protected]) wrote: > Mathieu, > I have another version of the kernel that I need to port LTT to. What > I was wondering is if I should take the latest version of what you've put > together. In particular will that have everything that went into the CSO > work, and is it stable enough. I was thinking I would just take the patch > that I had been working with, but it occurred to me that since I have to do > the work it might make sense to do it to the latest. I'd be happy to chat > about it if you have any advice and would prefer to talk otherwise letting > me in email is fine. > > -bob > OpenPGP public key: http://krystal.dyndns.org:8080/key/compudj.gpg Key fingerprint: 8CD5 52C3 8E3C 4140 715F BA06 3F25 A8FE 3BAE 9A68