This is the mail archive of the
systemtap@sourceware.org
mailing list for the systemtap project.
Re: kernel function probe overhead
- From: Josh Stone <jistone at redhat dot com>
- To: Aubrey Li <aubreylee at gmail dot com>, "Frank Ch. Eigler" <fche at redhat dot com>
- Cc: systemtap at sourceware dot org
- Date: Tue, 29 Nov 2016 11:20:59 -0800
- Subject: Re: kernel function probe overhead
- Authentication-results: sourceware.org; auth=none
- References: <CAGPKeUKWAvBv=SL_ebVkR3ZwCeGZoCQ1MEJ5xQi-VXuF=kt3Wg@mail.gmail.com> <y0m1sxzpj9i.fsf@fche.csb> <CAGPKeUJjf5qqxO2Gqo-_LzzaFbwQ=-wZ=sq0fwTzm=c+MEAtSw@mail.gmail.com> <y0mmvgjo1jl.fsf@fche.csb> <CAGPKeULiGUPMQ6e-YjSY3MTW+r=rjxYHYmoKEb1sU4DN+SL1Lg@mail.gmail.com>
On 11/28/2016 07:30 PM, Aubrey Li wrote:
>> Another solution would be to have per-thread global variables. Since
>> > the kernel doesn't help us out with this, we'd have to emulate it
>> > somehow, and for this to be wortwhile, it'd have to work faster than
>> > ordinary locked/shared array[tid()] accesses.
>> >
> Is this a suggestion for me? How to do this?
There are relevant RFEs for stap already:
per-cpu: https://sourceware.org/bugzilla/show_bug.cgi?id=10795
per-thread: https://sourceware.org/bugzilla/show_bug.cgi?id=10796
The latter was closed on the hunch that it wouldn't help much, but that
could still be explored if someone is interested.