Messages in this thread | | | Date | Fri, 16 Dec 2011 17:07:58 +0100 | From | Hans Rosenfeld <> | Subject | Re: [PATCH 0/9] rework of extended state handling, LWP support |
| |
On Mon, Dec 05, 2011 at 11:22:23AM +0100, Ingo Molnar wrote: > > * Hans Rosenfeld <hans.rosenfeld@amd.com> wrote: > > > These patches were built and tested against 3.1. The older RFC > > patches that have been lingering in tip/x86/xsave for the last > > few months should be removed. > > They have been lingering because of negative review feedback i > have given to you about LWP. I'm not convinced about the current > form of abstraction that this patch-set offers. > > See this past discussion from half a year ago: > > Subject: Re: [RFC v3 0/8] x86, xsave: rework of extended state handling, LWP support > > We can and should do better than that.
We had an intern, Benjamin Block, working on perf support for LWP until a few weeks ago. Because of the fundamental problem that it is not reasonably possible to allocate virtual memory for a process from a different process' context, the code only supports self-monitoring.
This allows a process to control LWP through the perf syscalls. Instead of using malloc() and the LLWPCB instruction itself, it can use the perf syscall to have the perf kernel code do it. It can also use perf to get at the raw LWP samples instead of just reading them from the LWP buffer.
I'll send you Benjamins code as an RFC patch set, so please take a look at it and tell me what you think about it. I admit I don't understand every detail of it yet as I've only recently started to work my way through it, and I have no prior knowledge about perf.
Hans
| |